International Journal of Environmental Research and Public Health Technical Note Digital Reconstitution of Road Traffic Accidents: A Flexible Methodology Relying on UAV Surveying and Complementary Strategies to Support Multiple Scenarios Luís Pádua 1,2, * , José Sousa 1 , Jakub Vanko 1 , Jonáš Hruška 1 , Telmo Adão 1,2 , Emanuel Peres 1,2 , António Sousa 1,2 and Joaquim J. Sousa 1,2 1 2 * Engineering Department, School of Science and Technology, University of Trás-os-Montes e Alto Douro, 5000-801 Vila Real, Portugal; jmsousa@utad.pt (J.S.); vanko.jakub@gmail.com (J.V.); jonash@utad.pt (J.H.); telmoadao@utad.pt (T.A.); eperes@utad.pt (E.P.); amrs@utad.pt (A.S.); jjsousa@utad.pt (J.J.S.) Centre for Robotics in Industry and Intelligent Systems (CRIIS), INESC Technology and Science (INESC-TEC), 4200-465 Porto, Portugal Correspondence: luispadua@utad.pt; Tel.: +351-25-9350-762 (ext. 4762) Received: 30 January 2020; Accepted: 10 March 2020; Published: 13 March 2020 Abstract: The reconstitution of road traffic accidents scenes is a contemporary and important issue, addressed both by private and public entities in different countries around the world. However, the task of collecting data on site is not generally focused on with the same orientation and relevance. Addressing this type of accident scenario requires a balance between two fundamental yet competing concerns: (1) information collecting, which is a thorough and lengthy process and (2) the need to allow traffic to flow again as quickly as possible. This technical note proposes a novel methodology that aims to support road traffic authorities/professionals in activities involving the collection of data/evidences of motor vehicle collision scenarios by exploring the potential of using low-cost, small-sized and light-weight unmanned aerial vehicles (UAV). A high number of experimental tests and evaluations were conducted in various working conditions and in cooperation with the Portuguese law enforcement authorities responsible for investigating road traffic accidents. The tests allowed for concluding that the proposed method gathers all the conditions to be adopted as a near future approach for reconstituting road traffic accidents and proved to be: faster, more rigorous and safer than the current manual methodologies used not only in Portugal but also in many countries worldwide. Keywords: road traffic accidents; reconstitution; unmanned aerial systems; photogrammetry 1. Introduction Regardless of the many road safety programs already implemented, undergoing or foreseen—e.g., the European Union (EU) is discussing their 5th Road Safety Action Programme for the 2020–2030 decade [1], intended to have a 50% reduction on both deaths and serious injuries between 2020 and 2030—related rates recorded worldwide are still significant. Indeed, the World Health Organisation (WHO) estimates that, annually, road traffic accidents kill around 1.25 million people and injure about 50 million more across the globe. In fact, in 2012, road traffic accidents were the leading cause of death between ages 15 and 29 years old [2]. Furthermore, the International Transport Forum (ITF), who gathers information from 40 countries in its 2017 Road Safety Annual Report [3], corroborates the aforementioned numbers related with casualties. Int. J. Environ. Res. Public Health 2020, 17, 1868; doi:10.3390/ijerph17061868 www.mdpi.com/journal/ijerph Int. J. Environ. Res. Public Health 2020, 17, 1868 2 of 24 Besides the social consequences, road traffic accidents also imply considerable economic impacts. Wijnen and Stipdonk [4] estimated the economic impacts as a percentage of countries gross domestic product (GDP): an average of 2.7% GDP to those they regarded as high-income countries and 2.2% GDP to low and middle- income countries. Examples of road traffic accidents economic costs in some countries were presented by the ITF [3]. This enormous economic impact weights down on families, insurance companies (e.g., damages), healthcare systems (both public and private), companies (e.g., production loss) and society as a whole [2,4]. Road safety is usually related with human behaviour, physical infrastructures and vehicles [5,6] (detailed information about road safety factors can be found in Rune and Vaa [7]). Countless media campaigns are made worldwide each year to raise awareness and try to reduce on-road risk behaviours. WHO has even released a mass media campaigns toolkit to help low- and middle-income countries develop their own initiatives [8]. Physical infrastructures design and planning, as well as vehicles active and passive safety measures or regular condition checking aim at giving a contribution to improve road safety [9,10]. When all else fails and road traffic accidents happen, it is crucial to carry out a crash scene reconstitution with the utmost rigor and objectivity, to determine the causes underlying the event. This information is the basis in which all of the accidents subsequent stages lean on. Whilst investigation, insurances, legal aspects and property damages are stages that directly depend on crash-site gathered information. This information is used to determine the causes of the event which can potentially have a valuable contribution in the aftermath, namely in the reduction of road traffic accidents. Indeed, an accurate crash scene reconstitution may allow the identification of design flaws in physical infrastructures design flaws and/or risk behaviours that support their mitigation through physical interventions or by raising awareness. Therefore, data collection at the crash site takes on a decisive importance [11]. Police forces are usually the first to arrive to a road traffic accident site. Ordinarily, their mission is first of all to assure victims (if any) proper medical assistance and then to secure the crash scene, preserving both property and evidences alike, followed by the traffic re-routing procedures. Afterwards, the investigation procedure starts by identifying and secure evidence, drawing sketches (e.g., vehicles positions and paths, visible traffic signs), take note of on-site conditions (e.g., visibility), take measures and sometimes, acquire photographs for future memory [12]. Besides the foremost objective to preserve life, restoring normal traffic flow to the affected roads is a prime directive, avoiding yet more social and economic impacts. In addition to the difficulties connected with the need of collecting a high amount of information on site, sometimes measurements taken from specific references cannot be met. For example, in what regards distances from the wheels of the vehicles, these should be measured from their centre and not from the edge of the tire. Moreover, this must be done from the same location, to the reference point. Under adverse weather conditions, there is also the risk of some evidences becoming less visible or even vanish. Furthermore, if pedestrians are involved in the accident, the information to be gathered increases considerably. Thus, data collection on site should be a simple, precise and swift procedure. However, currently it is mostly a manual procedure done by police forces worldwide. Although there might be slight differences from country to country, data collection methods used when road traffic accidents occur are very similar. They are based mainly in measurements done manually, using triangulations and parallel lines to produce the crash scene sketch: as such, both rigor and precision of a road traffic accident scene sketch depend mostly on the professionalism of the police forces. These difficulties can be overcome using small-sized, lightweight and cost-effective multi-rotor unmanned aerial vehicles (UAVs), capable of operating in fully automatic mode or by manual control, which constitute the technological core of Unmanned Aerial Systems (UAS)-based photogrammetric methodology proposed for road traffic accidents reconstitution. More than providing scientific documentation, this technical note intends to supply road traffic authorities and professionals with a set of procedures compiled in a UAV-based Int. J. Environ. Res. Public Health 2020, 17, 1868 3 of 24 methodology that aims to achieve accurate, efficient and cost-effective digital reconstruction and recording of motor vehicle collision events. 2. Background Recent photogrammetry methods were boosted by solid and fast pace developments carried out in digital image processing and computer vision. Some feature extraction algorithms—for example, scale invariant feature transform (SIFT) [13] and structure from motion (SfM) analysis [14].Due to its capability of generating wide and detailed digital data sets faithfully portraying real-world sites, photogrammetry has been gradually adopted in many areas of application [12]. One of them is the reconstruction and analysis of road traffic accidents in several projects and countries. A report provided by Arnold et al. [15] explores alternatives to the laborious and time-consuming traditional road traffic accident surveying methods. Close-range photogrammetry is highlighted. In spite of the resistance that is still noticed in adopting advanced techniques, many law enforcement entities around the globe admit the benefits of photogrammetry in both ease of use and time-saving factors, which has been leading to progressive changes in road traffic accidents surveying operations [16]. Zuraulis et al. [17] showed the relevance of photogrammetric systems for road traffic accident investigation, specifically by using tire yaw marks as evidences. This study suggests that the speed calculation methodology depends on tire curvature conditions, with lower errors for shortest chord lengths. Photogrammetric methods demonstrated to be useful for substantial accuracy improvement in speed determination: around 3% error was obtained when using curvatures shortest chord lengths, which is an acceptable value for forensic practices. However, some disadvantages regarding close-range photogrammetry have been noted [18]. With aerial photogrammetry, namely using Unmanned Aerial Systems (UAS)—that combine a UAV, sensors as payload and a ground station, which has software for flight management and planning [19]—both complexity and laborious road traffic accidents surveying operations dropped significantly, comparatively to the previously referred approach, relying on multi-positional photographs. This technological advancement turned UAV-based photogrammetry into a desirable precise and efficient tool for business-to-business companies and law enforcement entities dealing with road traffic accidents which has been also drawing the attention of researchers. Monadrone [20] used a UAV to acquire images from a simulated accident scene. The Royal Canadian Mounted Police and Pix4D company cooperated with the goal of comparing terrain surveying methods for road traffic accidents investigation—including laser 3D scanning—with UAV-based image acquisition [21]. The UAV-based photogrammetry reliability over traditional methods and its affordability comparatively to other precision instruments—e.g., laser scanners—were highlighted, along with other aspects such as view angles covering. A solution proposed by the Analist Group [22] resorts to terrain measurements and a UAV flight to obtain data for mosaicking and 3D model processing, using Pix4Dmapper Pro, both portraying different perspectives of a staged road traffic accident. Subsequently, measurements can be done in a computer aided design (CAD) software, that also supports simulating events sequences. Greg Gravesen, a specialist in road traffic accidents reconstitution who uses a fixed-wing UAV to create detailed representations from orthophoto mosaics are the base for the various simulations, as well as 3D meshes to create graphical hypothesis [23]. Crash scenes are then reconstructed and simulated using a dedicated software. Some research has also been done in this area by Su et al. [24] which proposes a mapping system that applies UAVs to rapidly set up a scene representation for road traffic accidents investigation, as an alternative to the traditional hand-drawn sketches. Diaz-Vilarino et al. [25] identified a non-solved question on runoff evaluation. They carried out a study to determine the influence of data quality obtained using UAV-based photogrammetry in the calculation of a mountain road run off, when compared to LiDAR data. A method to check the limits of using point clouds obtained from UAV photogrammetry relying on the D8 (flow direction) algorithm was proposed. Data acquisition of road traffic accidents comparing traditional methods with advanced techniques, namely using total Int. J. Environ. Res. Public Health 2020, 17, 1868 4 of 24 stations and photographic documentation that can also involve UAV missions, was carried out by Stáňa et al. [26]. Results analysing real occurrences as case studies pointed out both faster processing and quicker road clearance, as well as (almost) error-free measurements, when applying advanced techniques. A Sheriff of Colorado State (USA) testified for a report [27], sharing the same thoughts. X. Liu et al. [28] proposed a framework method that integrates UAVs to achieve 2D/3D accident scene reconstructions, highlighting the enhanced capabilities of those platforms to cover more angles without the need to interrupt traffic flow. The results seem to attest this kind of approaches to digitally record casualties in road traffic accidents. Concerned with mosaicking production, Y. Liu et al. [29] developed a method that includes image registration and fusion, as well as intelligent distance measurement based in Euclidian distances, a case study was used to demonstrate the whole pipeline. In Stevens Jr et al. [30], UAS were used for real-time monitoring of secondary crashes, including alternate routes monitoring. The report highlights UAS capabilities to deliver imagery and sensory data on-the-fly, their safety during missions, as well as payload mobility. Coarse 3D reconstructions were also presented to demonstrate mapping capabilities for future access, but quality measurements were not carried out. Mekker [31] evaluated some consumer grade UAS for photogrammetric traffic crash scene documentation. 3. Road Traffic Accidents Reconstitution 3.1. Current Approach The current procedures commonly used by police forces in road traffic accidents reconstitution relies on: a measuring tape and/or measuring wheel; a notepad, to draft the sketch; a pencil; chalk, to mark the pavement; and a camera for photographic documentation. Some constraints regarding the current data collection method are identified: (i) it involves a significant amount of time spent at the crash site by police officers that could otherwise be engaged in different tasks; (ii) it causes a disruption in regular traffic flow, with consequent economic and social impacts; (iii) it can potentially lead to losing valuable data, due to the lack of detail level in the sketch; and (iv) it exposes both police officers and civilians to unnecessary risks during the data collection process. Figure 1 depicts the current road traffic accidents data collection method applied to a simulated accident scenario. For this purpose, a simulation was conducted at the University of Trás-os-Montes e Alto Douro (UTAD), Vila Real, Portugal, with the cooperation of the Portuguese Public Safety Police (PSP) force from Vila Real, to ensure that the data collection process proceeded as if in a real road traffic accident scenario. Measurements done by PSP officers are in a local cartesian coordinate system: in this simulation, a lamp pole was used as the reference (x = 0.0 m; y = 0.0 m) and all measurements were referenced in relation to it. Moreover, a handmade sketch of the road traffic accident scenario was drawn into the notepad in situ (Figure 1c). For clarification purposes, the simulated road traffic accident was represented through a sketch drawn using an online platform (AccidentSketch.com) and is presented in Figure 1d. Besides measurements, some photographs are also usually required for documentation purposes. In this simple road traffic accident scenario, PSP took about 30 min to complete the data collection process. The handmade road traffic accident sketch (Figure 1c) was later used in the office to generate a digital version using a proprietary software. It was then included in the official accident report. This process took a few more hours of the police officers’ time, which could otherwise be used in more meaningful police work. 3.2. Proposed Approach Considering a set of requirements aligned with the cooperation of Portuguese police forces, some preliminary technical decisions regarding the selection of hardware (UAV, camera and base station) and software (UAV flight planner/manager and photogrammetric tools) were specified. Relying on such requirements and specifications, the outline of a novel methodology to digitally reconstruct and record road traffic accidents was proposed to complement or, eventually, supplant traditional procedures. simulated road traffic accident was represented through a sketch drawn using an online platform (AccidentSketch.com) and is presented in Figure 1d. Besides measurements, some photographs are also usually required for documentation purposes. In this simple road traffic accident scenario, PSP took about 30 minutes to complete the data collection process. The handmade road traffic accident sketch (Figure 1c) was later used in the office to generate a digital version using a proprietary Int.software. J. Environ. Res. Public Health 2020, 17,in 1868 of 24 It was then included the official accident report. This process took a few more hours5 of the police officers’ time, which could otherwise be used in more meaningful police work. Figure 1. Current data collecting method applied by the Portuguese Public Safety Police (PSP) of Vila Figure 1. Current data collecting method applied by the Portuguese Public Safety Police (PSP) of Vila Real (Portugal) in a simulated road traffic accident scene: (a) staged traffic accident scene; (b) police Real (Portugal) in a simulated road traffic accident scene: (a) staged traffic accident scene; (b) police officer using a measuring wheel to acquire data; (c) in situ sketch example; and (d) traffic accident officer using a measuring wheel to acquire data; (c) in situ sketch example; and (d) traffic accident scene scene sketch produced for clarification purposes by using an online platform. sketch produced for clarification purposes by using an online platform. 3.2.Requirements Proposed Approach 3.2.1. Definition Defining a set of requirements to a methodology that, from the outset, is intended not only to demonstrate its feasibility but also to effectively be deployed and employed in real road traffic accidents scenarios, can be a demanding task. As such, cooperation was sought from the Portuguese National Road Safety Authority (ANSR)—whose mission is the “planning and coordination at national level to support the Government policy on road safety, as well as the application of the law on road safety”—and both police forces that deal with road traffic accidents in Portugal: PSP and National Republican Guard (GNR). This cooperation resulted in the following set of requirements to which a UAS, to be used in road traffic accidents scenarios by police forces, must comply to be deemed suitable: (1) balanced relation between cost, reliability and precision; (2) compact, easy to manage, to transport and to assemble; (3) able to operate in both urban and nonurban contexts; (4) support both autonomous and manual flight modes; (5) high manoeuvrability and adaptability both in autonomous and in manual flight modes; (6) enable flights parameterisation to support a multiplicity of accident scenarios, namely regarding different types of trajectories, camera angle and images overlap; (7) autonomy to acquire data in typical road traffic accidents scenarios, using only one battery (assuming that most of road traffic accidents involves between two to three vehicles and that a common rotary-wing UAV can fly up to twenty minutes with only one battery, one flight is enough to cover the large majority of occurrences. Regarding multiple vehicle collisions in motorways, crash scenarios with highly dispersed debris or even involving explosions, the proposed methodology can be also addressed, but they may imply the use of more than one battery to fully cover the crash scenario); (8) integration of an accurate Global Navigation Satellite System (GNSS) receiver for positioning and imagery georeferencing; (9) support for RGB sensors that acquire high quality and high-resolution images up to 120 m altitude, which is the legal limit allowed for UAVs, according to European regulations (UAV flights are forbidden in some urban contexts, but as the proposed methodology will be applied by police forces, these situations must be foreseen in the legislation); and (10) allow for real-time access to the RGB sensor, through the ground control station, enabling the on-scene police officers to monitor and manage the data acquisition process and make an on-the-fly assessment about data being acquired. Regarding post-flight data processing, more specifically the photogrammetric processing, another set of specifications was created for selecting the photogrammetric tools: (1) support for photos positioning and alignment, aiming the global or local data georeferencing and feature extraction; (2) capability for data scaling operations; (3) intuitive and easy-to-use software; (4) support for accurate dense point clouds; (5) support interactive highly accurate measurements and positional control; and (6) outputs such as orthorectified GIS-based products and virtual models. 3.2.2. Hardware and Software Specification Considering the set of requirements defined for UASs, the process of selecting the equipment deemed suitable for the proposed methodology and for this study purpose started off by addressing aspects such as packaging, transportation and ease of handling. Among the available types, small-sized Int. J. Environ. Res. Public Health 2020, 17, 1868 6 of 24 rotary-wing UAVs superseded fixed-wing UAVs due to their ability to perform vertical take-off and landing (VTOL) and to maintain a stationary position during flight. The authors experience in working and operating with different UAVs also weighed in when selecting the two off-the-shelf UAV models for this study: DJI Phantom 4 and its predecessor, DJI Phantom 3 Standard (DJI TECHNOLOGY CO., LTD, Shenzhen, China). For mission planning and execution, two software tools were used: DJI GO (DJI TECHNOLOGY CO., LTD, Shenzhen, China) and Pix4Dcapture (Pix4D SA, Lausanne, Switzerland). While the former stands as a valid option to test UAV configurations, the latter is suitable for preparing and executing mission flights that can be carried out manually or autonomously—in simple or double grid mode—and also for selecting which area to cover, the flight height and imagery overlap. To complement UAV-based imagery, ground-perspective images can be collected by using ground cameras, namely smartphones belonging to a widely spread class of cost-effective consumer-grade technological devices, with built-in GPS capabilities and equipped with fairly acceptable quality cameras (12 MP). With these, police officers are able to deal with less accessible areas in the crash scene that cannot be tackled with—only or at all—using UAVs, due to the presence of obstacles (e.g., dense vegetation, poles, cables or buildings). After acquiring data from road traffic accidents scenarios, a photogrammetric processing stage occurs. As such, notwithstanding several fully compliant solutions (some of them opensource and/or free to use), Pix4Dmapper Pro (Pix4D SA, Lausanne, Switzerland) was selected for this study. The decisive factors for this choice were the research team significant experience in using this software and an available license. The resulting digital products from this pipeline represent a realistic and complete set of elements that can help to reconstitute road traffic accidents scenarios. 3.2.3. Environmental Conditions Variability UAV-based photogrammetric approaches for road traffic accidents reconstitution are usually equipped with one RGB camera to acquire images from the scene and GNSS, which besides georeferencing each image, enables autonomous flights throughout pre-established set of waypoints (i.e., mission plans). As such, human intervention can be minimised during the data acquisition process, thus reducing the potential for errors. However, some context conditions might significantly influence this process, namely (1) the presence of unmovable obstacles that partially or completely prevent UAVs access to the crash scenario; and (2) adverse light conditions, often found with some environmental phenomena and/or at night. Acquired images are then processed by using a photogrammetric software to accurately reconstitute the traffic accident scenario. 3.2.4. Methodology Overview When a road traffic accident happens and assuming that police forces are called to the scene, the proposed methodology starts of by having officers evaluate the crash scenario (e.g., area of interest, number of vehicles involved, dispersion of potential elements of interest) and the context conditions (e.g., environmental and light conditions and the presence of obstacles). Indeed, four road traffic accident scenarios are foreseen, each with its corresponding imagery acquisition process, as summarised in Table 1. During the imagery acquisition process, a small number of distances between reference points can be measured to ascertain the road traffic accident scenario reconstitution quality and also for calibration purposes: bars with well-known dimensions can be used (placing them in the area of interest in concurrent directions), measuring tape (or wheel) or by employing a GNSS receiver. Although it is worth mentioning that if the objective is to reconstitute a road traffic accident scenario circumscribed to a relatively small area (a few square meters) by using a UAV equipped with a precise GNSS receiver (with errors around a few centimetres), then the distance measurement between reference points Int. J. Environ. Res. Public Health 2020, 17, 1868 7 of 24 may be unnecessary, due 2020, to photogrammetric software capabilities in estimating 3D positions 7with Int. J. Environ. Res. Public Health 17 of 24 acceptable accuracy. Artificial light sources can be setup around the area of interest. As road traffic accident scenarios andofrespective Table 1. Foreseen Adverse light proposedanmethodology alternative/complement, whenever the use UAVs is imagery possible, acquisition processes. UAV: unmanned aerial vehicle. conditions portable lightweight light sources can be mounted on the UAV, ensuring the camera is neither occluded or overexposed. Foreseen Road Traffic Accident Scenarios Absence of obstacles (ideal situation) Imagery Acquisition Process A UAV can carry out a fully autonomous flight, planned with proper software Image acquisition can be complemented either using manual flight (for wider During imagery acquisition Presencethe of some obstacles, restricting process, a small number of distances between reference points areas) or by using a ground camera, such as a smartphone (for confined areas UAV access to (at least) one perspective can be measured to ascertain the road traffic accident scenario quality and also for of a few squared meters,reconstitution orbital style acquisition) calibration purposes: bars with well-known can them Image dimensions acquisition is done onlybe by used using a (placing ground camera, suchin as athe area of Presence of obstacles that completely smartphone (more suitable for confined areas of a few squared meters, orbital interest in concurrent measuring tape (or wheel) or by employing a GNSS receiver. inhibit the use directions), of UAVs style acquisition) Although it is worth mentioning that if the objective is to reconstitute a road traffic accident scenario Artificial light sources can be setup around the area of interest. As an alternative/complement, whenever the of UAVs possible, portable with a circumscribed to a relatively small area (a few square meters) byuse using a is UAV equipped Adverse light conditions lightweight light sources can be mounted on the UAV, ensuring the camera is precise GNSS receiver (with errors around a few neither centimetres), then the distance measurement occluded or overexposed. between reference points may be unnecessary, due to photogrammetric software capabilities in estimating 3D positions with acceptable accuracy. Lastly, the acquired data concerning the road traffic accident scenario must be processed using a Lastly, the acquired data concerning the road traffic accident scenario must be processed using photogrammetric software to obtain an orthophoto mosaic and a virtual model reconstruction, therefore a photogrammetric software to obtain an orthophoto mosaic and a virtual model reconstruction, digitally documenting the road traffic accident. The crash scenario reconstitution is represented by therefore digitally documenting the road traffic accident. The crash scenario reconstitution is a georeferenced 3D model or point cloud, in which precise measurements can be taken, edited or represented by a georeferenced 3D model or point cloud, in which precise measurements can be complemented at any point in time, through the use of specific software capable of dealing with taken, edited or complemented at any point in time, through the use of specific software capable of Geographic Information Systems (GIS) data and/or Computer Aided Design (CAD) models. Figure 2 dealing with Geographic Information Systems (GIS) data and/or Computer Aided Design (CAD) presents the proposed methodology functional diagram. models. Figure 2 presents the proposed methodology functional diagram. Figure2.2.Unmanned Unmanned aerial aerial vehicle vehicle (UAV)-based (UAV)-based road Figure road traffic traffic accidents accidents reconstitution reconstitutionmethodology methodology functional diagram. functional diagram. 4. Experimental Tests: Tests: Fine-Tuning Fine-Tuning Parameterisation Parameterisation for for the TheProposed ProposedMethodology Methodology 4.1. Experimental Scenario Scenario 4.1. Experimental A most suitable context conditions (clear sky,sky, obstacles-free A set set of ofpreliminary preliminarytests, tests,considering consideringthe the most suitable context conditions (clear obstacleswide-open area),area), were performed with various heights heights and distinct in angles order toinassess free wide-open were performed with various and camera distinctangles camera orderthe to proposed methodology feasibility. These tests were also used to determine flight parameters default assess the proposed methodology feasibility. These tests were also used to determine flight parameters default settings such as flight duration, most suitable number of acquired images, processing time, Ground Sample Distance (GSD) and measurements precision. More specifically two experimental tests sets were completed: one to optimise flights height and duration, as well as the resulting digital products overall quality, including resolution and 3D model detail; another to Int. J. Environ. Res. Public Health 2020, 17, 1868 8 of 24 settings such as flight duration, most suitable number of acquired images, processing time, Ground Sample Distance (GSD) and measurements precision. More specifically two experimental tests sets were completed: one to optimise flights height and duration, as well as the resulting digital products Int. J. Environ. Res. Public Health 2020, 17 8 of 24 overall quality, including resolution and 3D model detail; another to compare both precision and executionboth timeprecision of the current data collection and the proposed methodology, a road compare and execution time ofmethod the current data collection method and within the proposed traffic accidentwithin scenario context. As aaccident result, near-optimal parameter combinations were determined methodology, a road traffic scenario context. As a result, near-optimal parameter for future reference. combinations were determined for future reference. 4.2. Flight Height and Camera Angle 4.2. Flight Height and Camera Angle Both flight height and camera angle are important parameters inasmuch as they influence Both flight height and camera angle are important parameters inasmuch as they influence orthorectified and virtual model outcomes obtained through photogrammetric processing of imagery orthorectified and virtual model outcomes obtained through photogrammetric processing of imagery acquired in a given area of interest. Regarding camera angle, nadiral images would allow for obtaining acquired in a given area of interest. Regarding camera angle, nadiral images would allow for the best orthorectified mosaics. However, it is a less favourable situation for obtaining 3D models obtaining the best orthorectified mosaics. However, it is a less favourable situation for obtaining 3D due to a poorer perspective coverage. With regard to flight height, it stands to reason that lower models due to a poorer perspective coverage. With regard to flight height, it stands to reason that flight heights allow for a higher spatial resolution. This can be important, to detect vehicles damage lower flight heights allow for a higher spatial resolution. This can be important, to detect vehicles and existent debris. Nevertheless, in road traffic accident scenarios, namely those in urban contexts, damage and existent debris. Nevertheless, in road traffic accident scenarios, namely those in urban lower heights can induce more operational constraints and hamper the detection of similar features in contexts, lower heights can induce more operational constraints and hamper the detection of similar extreme situations, with a negative impact on the photogrammetric processing. Experimental tests features in extreme situations, with a negative impact on the photogrammetric processing. were conducted to find the best possible compromise between these two parameters, aiming to obtain Experimental tests were conducted to find the best possible compromise between these two the most suitable outcomes from the photogrammetric processing. parameters, aiming to obtain the most suitable outcomes from the photogrammetric processing. To test and assess combinations between these two parameters, several flight missions were To test and assess combinations between these two parameters, several flight missions were carried out under ideal circumstances (clear sky sunny days with weak or non-existent wind, within an carried out under ideal circumstances (clear sky sunny days with weak or non-existent wind, within obstacle-free wide-open area), establishing accidents scenarios in controlled conditions, as presented in an obstacle-free wide-open area), establishing accidents scenarios in controlled conditions, as Figure 3. presented in Figure 3. Figure 3. Simulated possible compromise compromise between parameters Figure 3. Simulated scenario scenario to to determine determine the the best best possible between flight flight parameters for traffic accidents reconstitution (a). The area considered for measurements was of 67 m22,, which for traffic accidents reconstitution (a). The area considered for measurements was of 67 m which corresponds to the blue dashed polygon (b), five measuring points are highlighted (A—E). corresponds to the blue dashed polygon (b), five measuring points are highlighted (A—E). Flights height varied between 5 m and 100 m, with 5 m increments. Flight Flight missions missions were were carried out manually for both the 5 m and 10 m heights. The purpose was to simulate a situation The purpose was to simulate a situation where where there there were eventual obstacles present, while maintaining aa clear clear line-of-sight line-of-sight with withthe theUAV. UAV. The remaining flights were all done in fully autonomous mode, using the same mission plan in Pix4Dcapture. The influence of different perspectives was evaluated by carrying out both simple and double grid flight missions. missions. With With regard regard to to camera camera angle, angle, itit stands stands to to reason reason that that it must be sharper in lower height flights and wider (until nadiral position) in higher height flights, to be able to capture accident scene ◦ to ◦ inin elements. It varied varied from from30 30° to6060° flights where height equal or lesser mfrom and flights where thethe height waswas equal to ortolesser thanthan 10 m10 and ◦ ◦ from 90°the forremaining the remaining flights. To assess positional and model generated scaling, 60 to60° 90 tofor flights. To assess bothboth positional rigorrigor and model generated scaling, two two perpendicularly arranged metal bars, with a known length (1 m), were placed within the perpendicularly arranged metal bars, with a known length (1 m), were placed within the simulated simulated traffic accidentFive scenario. Five points—from A to E—were marked in theusing pavement road trafficroad accident scenario. points—from A to E—were marked in the pavement chalk, using chalk, for validation purposes (Figure 3b). for validation purposes (Figure 3b). Table 2 presents data regarding the most representative combinations of the following parameters: type of flight mission—manual or autonomous (simple/double grid), flight height, camera angle, number of acquired images, flight mission duration, GSD and photogrammetric processing duration. As the flight height decreased, flight duration increased, along with the number of acquired images and the GSD. Furthermore, the number of acquired images demanded more computational resources during the automatic photogrammetric processing (therefore excluding Int. J. Environ. Res. Public Health 2020, 17, 1868 9 of 24 Table 2 presents data regarding the most representative combinations of the following parameters: type of flight mission—manual or autonomous (simple/double grid), flight height, camera angle, number of acquired images, flight mission duration, GSD and photogrammetric processing duration. As the flight height decreased, flight duration increased, along with the number of acquired images and the GSD. Furthermore, the number of acquired images demanded more computational resources during the automatic photogrammetric processing (therefore excluding manual operations, such as scale constraints handling and measurements determination), which increased its duration. Table 2. Most representative combinations of flight parameters and photogrammetric processing duration (PPD), considering the best possible compromise between flight height and camera angle. Flight typology as simple grid (SG) or double grid (DG); GSD: ground sampling distance. Height (m) Camera Angle Flight Typology Number of Images Flight Duration (min:s) GSD (cm) PPD (min:s) 40 40 40 30 30 30 15 15 15 10 90 80 65 90 80 65 90 80 65 30–60 SG DG DG SG DG DG SG DG DG Manual 18 15 36 18 36 36 32 72 76 76 3:52 2:27 3:39 3:35 3:27 3:27 4:34 4:58 4:56 4:46 1.70 1.54 1.81 1.21 1.08 1.33 0.65 0.59 0.72 0.34 3:32 4:33 9:25 3:03 11:11 7:43 4:36 21:59 14:02 30:18 Tape measurements made between the five points (from A to E) marked in the pavement of the road traffic accident scenario (Figure 3b) were used as reference to compare to those extracted from the photogrammetric-based virtual model, generated by Pix4Dmapper Pro. This allowed for both assessing and validating the proposed method accuracy. Table 3 presents the most representative results obtained from the different flight missions. Overall, lower flight heights, and therefore higher spatial resolutions, enabled more accurate measurements. The most accurate result was achieved by a double grid flight, carried out at 15 m height with the camera angle set to 65◦ . Table 3. Comparison between the reference measurements and those most representative computed from photogrammetric-based virtual models obtained from flight missions. Combinations of different flight parameters, such as flight mode—manual or autonomous (single/double grid)—flight height and camera angle are presented for each flight mission. The most accurate result is highlighted in bold. Measurement Measuring tape 40 m SG (90◦ ) 30 m SG (90◦ ) 15 m SG (90◦ ) 40 m DG (65◦ ) 30 m DG (65◦ ) 15m DG (65◦ ) 40 m DG (80◦ ) 30 m DG (80◦ ) 15 m DG (80◦ ) Manual Distance (cm) AB CE DE AC 286 292 ± 2 288 ± 2 286 ± 1 289 ± 2 284 ± 1 286 ± 1 285 ± 2 283 ± 1 286 ± 1 287 ± 1 165 169 ± 1 166 ± 2 165 ± 1 166 ± 1 165 ± 2 165 ± 1 164 ± 2 163 ± 1 164 ± 1 166 ± 1 93 95 ± 2 94 ± 2 93 ± 1 94 ± 1 93 ± 1 93 ± 1 92 ± 1 92 ± 1 93 ± 1 94 ± 1 249 255 ± 1 250 ± 1 248 ± 1 250 ± 2 247 ± 1 249 ± 1 247 ± 3 246 ± 1 249 ± 1 251 ± 1 Int. J. Environ. Res. Public Health 2020, 17, 1868 10 of 24 4.3. Precision: Can a UAV-Based Photogrammetric Approach Be Comparable? Another road traffic accident was simulated to assess the precision of photogrammetric-based outcomes, by comparison, with a GNSS receiver and a total station. Validation was achieved using measurements taken with a measuring tape as ground-truth. Figure 4a presents the 30 × 30 m accident Int. J. Environ. Public Healthwere 2020, considered 17 10 of 24 scene, whereRes. eight points (A, B, C, D, E, F, G and H) and nine distances were measured. Furthermore, the time required to apply each measuring procedure—proposed method, total station, measured. Furthermore, the time required to apply each measuring procedure—proposed method, GNSS receiver and tape—was also measured. total station, GNSS receiver and tape—was also measured. Figure 4. 4. Assessment Assessment of of different different measuring measuring methods methods in in aa car car accident accident scenario: scenario: (a) (a) aerial aerial view view of of the the Figure road traffic accident scenario, with the eight points (A–H) considered for comparing the measuring road traffic accident scenario, with the eight points (A–H) considered for comparing the measuring methods; measuring measuring distances distances using using (b) (b) the the GNSS GNSS receiver; receiver;and and(c) (c)the thetotal totalstation. station. methods; Current procedures procedures that that are are commonly commonly used used in in road road traffic traffic accidents’ accidents’ reconstitution reconstitution by by police police Current forces are are based based on on aa measuring measuring tape tape (or (or wheel). wheel). As such, measurements measurements acquired acquired using using this this method method forces were considered considered as as reference reference or or ground-truth. ground-truth. The nine distances (AB, BC, CD, EF, EF, GH, GH, AH, AH, BG, BG, CF CF were and DE) DE)were weremeasured measured approximately three minutes. Regarding GNSS, measurements were and in in approximately three minutes. Regarding GNSS, measurements were taken taken in approximately eight minutes with a Topcon HiPer II GNSS receiver (Topcon Corporation, in approximately eight minutes with a Topcon HiPer II GNSS receiver (Topcon Corporation, Tokyo, Tokyo, while Japan), while ensuring a positional 2 and 4 cm, in kinematic for Japan), ensuring a positional precisionprecision between between 2 and 4 cm, in kinematic mode. Asmode. for theAs total the total station, a couple points of additional points with known coordinates considered for station, a couple of additional with known coordinates were considered forwere orientation purposes, orientation purposes, besides the existing eight. Using a Topcon GPT-3005N equipment, the entire besides the existing eight. Using a Topcon GPT-3005N equipment, the entire process took about nine process took about nine minutes and 30 seconds (equipment configuration and data acquisition). The minutes and 30 s (equipment configuration and data acquisition). The proposed UAV-based method proposed UAV-based method was applied in based on knowledge in the previous section was applied based on knowledge presented the previous sectionpresented and planned using Pix4Dcapture: and planned using Pix4Dcapture: an obstacle-free accident scene forwith an autonomous flight an obstacle-free accident scene allowed for an autonomous flight at 15allowed m height, a 65◦ camera angle at 1575% m height, a 65°Acamera and 75% images overlap. A total of 423images were and imageswith overlap. total ofangle 42 images were acquired in approximately min and 50 s.acquired Images in approximately 3 minutes and 50 seconds. Images were processed in approximately 2 minutes using were processed in approximately 2 min using Pix4Dmapper Pro running on a computer equipped Pix4Dmapper on CPU a computer equipped with a 2.6 GHz Intel i7-4720HQ CPU16 (Intel with a 2.6 GHz Pro Intelrunning i7-4720HQ (Intel Corporation, California, United States of America), GB Corporation, States of America), 16 GB memory (1600 MHz) and a NVidia GeForce memory (1600California, MHz) andUnited a NVidia GeForce GTX 970M (Nvidia Corporation, California, United States GTX 970M (Nvidia Corporation, California, States of4America) GPU. The resulting mosaic of America) GPU. The resulting mosaic had a United 0.55 GSD. Table presents the measurements acquired hadeach a 0.55 GSD. Table 4 presents the measurements acquired by each addressed method. by addressed method. GNSS presented the most accentuated deviations regarding ground-truth. When compared with Table 4. Comparison betweenmeasurements measurements taken measuring tape, the proposed UAV-based the (high-precision) total station, takenusing usinga the proposed UAV-based method presented photogrammetric approach, a total station and a GNSS receiver. The difference to the reference a quite similar accuracy, but with better time regarding operational performance. Moreover, operating measurements (measuring tape) to the one taken by the other equipment is also provided. a UAV involves less effort due to the autonomous flight capabilities. Another advantage when using the proposed approach it provides more (cm) data forTotal road traffic accidents reconstitution—through Distanceis that Tape (cm) GNSS station (cm) UAV (cm) acquired aerial imagery conversion into orthophoto mosaics, dense point clouds AB 257 259 (−2) 258 (−1) 258and (−1)textured virtual models—that canBC be achieved180 by resorting179 to user-friendly photogrammetric software (1) 180 (0) 181 (−1) tools. CD 268 267 (1) 269 (−1) 269 (−1) EF 264 262 (2) 264 (0) 265 (−1) GH 264 263 (1) 264 (0) 264 (0) AH 377 380 (−3) 377 (0) 378 (−1) BG 440 443 (−3) 440 (0) 441 (−1) CF 368 370 (−2) 369 (−1) 369 (−1) DE 375 377 (−2) 376 (−1) 376 (−1) GNSS presented the most accentuated deviations regarding ground-truth. When compared with Int. J. Environ. Res. Public Health 2020, 17, 1868 11 of 24 Table 4. Comparison between measurements taken using a measuring tape, the proposed UAV-based photogrammetric approach, a total station and a GNSS receiver. The difference to the reference measurements (measuring tape) to the one taken by the other equipment is also provided. Distance Tape (cm) AB 257 BC 180 CD 268 EF 264 GH 264 AH 377 BG 440 CF 368 Int. J. Environ. Res. Public Health DE 375 2020, 17 GNSS (cm) Total station (cm) UAV (cm) 259 (−2) 179 (1) 267 (1) 262 (2) 263 (1) 380 (−3) 443 (−3) 370 (−2) 377 (−2) 258 (−1) 180 (0) 269 (−1) 264 (0) 264 (0) 377 (0) 440 (0) 369 (−1) 376 (−1) 258 (−1) 181 (−1) 269 (−1) 265 (−1) 264 (0) 378 (−1) 441 (−1) 369 (−1) of 24 376 11 (−1) Moreover, operating a UAV involves less effort due to the autonomous flight capabilities. Another 5. Experimental Tests: Scenarios advantage when Simulated using the proposed approach is that it provides more data for road traffic accidents reconstitution—through acquired aerial imagery conversion into orthophoto mosaics, dense point The proposed proved to be a can validbeapproach reconstitution of road traffic clouds andmethodology textured virtual models—that achieved for by the resorting to user-friendly photogrammetric software tools. accidents scenarios occurring in obstacles-free wide-open areas, under clear sky sunny days. However, to be deployed and used by police forces under real scenarios, this methodology needed to be 5. Experimental Tests: Simulated Scenarios assessed in less favourable conditions. Obstacles that can commonly be found in urban and non-urban proposed methodology proved to be a valid approach for the reconstitution of road traffic environments,The such as aerial electrical cables, buildings, walls, vegetation and trees, were considered accidents scenarios occurring in obstacles-free wide-open areas, under clear sky sunny days. in simulated crash scenarios. Low light conditions or even night-time environments were also However, to be deployed and used by police forces under real scenarios, this methodology needed consideredtoin order toinfully evaluate conditions. the proposed methodology. Considering findings documented be assessed less favourable Obstacles that can commonly be foundthe in urban and nonurban environments, such as aerial electrical cables, buildings, walls, vegetation and trees, were in Section 4.2, all flights presented hereinafter had images overlap between 70% and 80%. All considered in simulated crash scenarios. Low light conditions or even night-time environments measurements of the tests performed in this section were taken using a measuring tape.were 5.1. also considered in order to fully evaluate the proposed methodology. Considering the findings documented in Section 4.2, all flights presented hereinafter had images overlap between 70% and Scenario 1: Poles and Electrical/Communications Aerial Cables 80%. All measurements of the tests performed in this section were taken using a measuring tape. The first scenario was simulated in an area with electrical and telecommunications poles, which 5.1. Scenario 1: Poles and Electrical/Communications Aerial Cables are very common obstacles in both rural and urban environments. As can be observed in Figure 5, first scenario was simulated in an area with electrical and telecommunications poles, which their presenceThe does not favour UAV flights at lower flight heights using autonomous flight modes. are very common obstacles in both rural and urban environments. As can be observed in Figure 5, Under these conditions, a manual flight mode—providing full flight control to the operator and thus their presence does not favour UAV flights at lower flight heights using autonomous flight modes. the responsibility to avoid/bypass obstacles—seems to be the data Under these conditions, a manual flight mode—providing fullmost flight reasonable control to theoption operatorfor andacquiring thus the responsibility to avoid/bypass obstacles—seems to be the most reasonable option for acquiring in a safe way for both police officers and equipment. In this simulated scenario the camera gimbal was in a safetoway for both police officers and equipment. thistosimulated theoverlap camera between manually data controlled adopt several orientations. The goal In was ensure ascenario suitable gimbal was manually controlled to adopt several orientations. The goal was to ensure a suitable images and to properly cover all the scene during on-the-fly data acquisition, towards a faithful and overlap between images and to properly cover all the scene during on-the-fly data acquisition, highly detailed reconstitution. A total3D of reconstitution. 119 images were in about four minutes. towards3D a faithful and highly detailed A totalacquired of 119 images were acquired in about Figure 5 four minutes. Figure 5 presents an aerial viewscenario, of the simulated scenario,points with six(A, control presents an aerial view of the simulated crash with crash six control B, C,points D, E and F). (A, B, C, D, E and F). Figure 5. Road traffic accident simulation in a context with poles and aerial cables: (a) aerial view of Figure 5. the Road traffic accident simulation in a context with poles and aerial cables: (a) aerial view road traffic accident scenario, where six control points (A–F) were marked to compare results of the road trafficthe accident where six control points (A–F) were to compare results between proposedscenario, approach and ground-truth measurements; (b) and (c) marked different perspectives of the accident scene, whereand the poles and aerial cables are visible. (b) and (c) different perspectives of between the proposed approach ground-truth measurements; the accident scene, where the poles and aerial cables are visible. A subset of measurements—AB, CD and EF—for comparison with the resulting point cloud is recorded in Table 5. The maximum error is of 2 cm (0.44%) was observed in AB. Table 5. Comparison between ground-truth measurements and the UAV-based photogrammetric approach relative to road traffic accident in a context with poles and aerial cables. Errors associated with the UAV-based measurements were calculated using Pix4D. Int. J. Environ. Res. Public Health 2020, 17, 1868 12 of 24 A subset of measurements—AB, CD and EF—for comparison with the resulting point cloud is recorded in Table 5. The maximum error is of 2 cm (0.44%) was observed in AB. Table 5. Comparison between ground-truth measurements and the UAV-based photogrammetric approach relative to road traffic accident in a context with poles and aerial cables. Errors associated with theRes. UAV-based measurements were calculated using Pix4D. Int. J. Environ. Public Health 2020, 17 12 of 24 Distance Distance AB AB CD CD EF EF Tape (cm) Tape (cm) 451 451 117 117 42 42 Point Cloud (cm) Point Cloud (cm) 453 ± 453 4 ±4 117 ± 1 117 ± 142 ± 1 42 ± 1 5.2. Scenario 2: Presence of Trees with Large Canopies 5.2. Scenario 2: Presence of Trees with Large Canopies Another common situation is the existence of natural obstacles in a road traffic accident scenario. Another common situation is the existence of natural obstacles in a road traffic accident scenario. A second simulated scenario was set in such a way that large trees obstructed portions of the involved A second simulated scenario was set in such a way that large trees obstructed portions of the involved vehicles and other possible elements of interest. Figure 6 depicts both the road traffic accident scenario vehicles and other possible elements of interest. Figure 6 depicts both the road traffic accident (Figure 6a) and the 3D representations from the photogrammetric processing (Figure 6b,c). In the latter scenario (Figure 6a) and the 3D representations from the photogrammetric processing (Figure 6b and it is clear that the crash scene portions obstructed by the trees canopy cannot be reconstituted by solely c). In the latter it is clear that the crash scene portions obstructed by the trees canopy cannot be using aerial images. Regarding UAV imagery, 43 images were obtained with approximately 2 min reconstituted by solely using aerial images. Regarding UAV imagery, 43 images were obtained with double-grid autonomous flight, at 30 m height and with a 70◦ camera angle. approximately 2 minutes double-grid autonomous flight, at 30 m height and with a 70° camera angle. Figure 6. 6. Road Roadtraffic trafficaccident accident simulation in which partially obstruct the involved vehicles. Figure simulation in which treestrees partially obstruct the involved vehicles. Aerial Aerial view of the scene and the six control points marked for measurement (A—F) (a); generated view of the scene and the six control points marked for measurement (A—F) (a); generated dense point dense (b); point cloud (b);model and the 3D model resulting from proposed methodology (c). cloud and the 3D resulting from applying theapplying proposedthe methodology (c). Table 6 presents presents aa subset subset of of results—AB, results—AB, BD, BD, CD, CD, AC, AC, EF—comparing EF—comparing the ground-truth measurements to the ones obtained from the resulting point cloud. All distances are very similar in both methods, with a maximum error of 1 cm cm occurring occurring in in all all measurements. measurements. Table 6.6. Comparison Comparisonbetween betweenground-truth ground-truth measurements taken the measurements obtained Table measurements taken andand the measurements obtained from the UAV-based point cloud a simulated road traffic where trees obstructs part the involved from the UAV-based point in cloud in a simulated roadaccident traffic accident where trees obstructs part the vehicles. associated with the UAV-based measurements were calculated using Pix4D. involved Errors vehicles. Errors associated with the UAV-based measurements were calculated using Pix4D. Distance Distance Tape (cm) Point Point CloudCloud (cm) (cm) Tape (cm) 69 69 ± 269 ± 2 AB AB 69 86 87 ± 187 ± 1 BD BD 86 CD CD 77 77 78 ± 278 ± 2 AC AC 122 122 123 ± 123 2 ±2 EF 79 80 ±1 EF 79 80 ±1 However, a complete crash scene reconstitution was not possible, due to the obstruction caused However, a complete reconstitution wasuseless not possible, caused by the trees canopy, which crash led toscene the aerial images being for thatdue parttoofthe theobstruction scenario. As such, by the trees canopy, which led to the aerial images being useless for that part of the scenario. As such, digital outcomes from the proposed methodology are somewhat incomplete, as can be seen in the 3D digital outcomes from the proposed methodology are somewhat incomplete, as can be seen in the model (Figure 6c). To overcome this issue and as mentioned in Section 4.2, ground-based image acquisition can be integrated with UAV-based imagery, thus enabling a full and accurate crash scenario reconstitution when parts are obstructed and not visible in the aerial imagery. 5.3. Scenario 3: Presence of Dense Vegetation Int. J. Environ. Res. Public Health 2020, 17, 1868 13 of 24 3D model (Figure 6c). To overcome this issue and as mentioned in Section 4.2, ground-based image acquisition can beRes. integrated UAV-based imagery, thus enabling a full and accurate crash Int. J. Environ. Public Health with 2020, 17 13 ofscenario 24 reconstitution when parts are obstructed and not visible in the aerial imagery. imagery acquisition was carried out in a manual flight that took approximately 4 minutes and 30 5.3. Scenario 3: atPresence of Dense seconds, a variable height Vegetation between 5 to 10 m and with a 70° camera angle, enabling a complete accident scene reconstitution. This third simulated road traffic accident presents a scenario where a significant portion of a Before the proposed methodology processing stage, the vegetation was removed by defining a vehiclesub-region and possible otherfocused elements of interest are obstructed byofdense vegetation (Figure 7a). This of interest on the vehicle (Figure 7b). A total 50 aerial images were acquired situation blocks the aerial view in part of the scene. Images acquired from the ground, using a using the UAV, which enabled the computation of a dense point cloud (Figure 7d). As the dense vegetation prevented thea reconstitution of one the vehicles sides, to document missingUAV-based part, smartphone equipped with 13 MP camera, wereofthus combined with the aerialthat imagery. 17 images from the obstructed areain were acquiredflight using that the smartphone camera—this4 process took imagery acquisition was carried out a manual took approximately min and 30 s, at ◦ approximately three minutes—and used to complement the aerial imagery (Figure 7e). Lastly, three a variable height between 5 to 10 m and with a 70 camera angle, enabling a complete accident measuring points (Figure 7f) were marked on the ground (A–C), to compare the reference scene reconstitution. measurements and the ones obtained from the resulting point cloud. Figure 7. Road traffic accident simulation with one vehicle in a dense vegetation context: (a) vehicle Figure 7. Road traffic accident simulation with one vehicle in a dense vegetation context: (a) vehicle aerial view; (b) point cloud vegetation filtering; (c) dense point cloud combining both ground and aerial view; (b) point cloud vegetation filtering; (c) dense point cloud combining both ground and aerial aerial imagery; (d) resulting UAV-based point cloud; (e) resulting 3D model of the vehicle side imagery; (d) resulting UAV-based point cloud; (e) resulting 3D model side obstructed by obstructed by vegetation, using ground-based imagery; and (f) partofofthe thevehicle resulting orthophoto vegetation, using ground-based imagery; points and (f)(A–C) part used of the mosaic, highlighting mosaic, highlighting in situ measuring forresulting precision orthophoto assessment purposes. in situ measuring points (A–C) used for precision assessment purposes. The comparison between ground-truth measurements (AB 209 cm and AC 308 cm) with the ones obtained the finalmethodology dense point cloud (AB 210 ±stage, 1 cm and 309 ± 1 cm), showed, again, 1 Before thefrom proposed processing the AC vegetation was removed byonly defining a cm discrepancy both7b). ground andofaerial imagery can were be combined to using sub-region of interest between focused them, on theproving vehicle that (Figure A total 50 aerial images acquired obtain a precise roadthe traffic accident scenario reconstitution using(Figure the proposed methodology. the UAV, which enabled computation of a dense point cloud 7d). As the dense vegetation prevented the reconstitution of one of the vehicles sides, to document that missing part, 17 images from 5.4. Scenario 4: Presence of Buildings the obstructed area were acquired using the smartphone camera—this process took approximately This fourthused scenario simulates a the roadaerial trafficimagery accident(Figure in which least one of measuring the involvedpoints three minutes—and to complement 7e).atLastly, three vehicles is fairly close to a building, thus having one of its sides obstructed to autonomous aerial (Figure 7f) were marked on the ground (A–C), to compare the reference measurements and the ones imagery acquisition. The aim is to depict a common situation, especially in urban contexts. Regarding obtained from the resulting point cloud. UAV imagery, 40 images were obtained with an approximately 4 minutes and 30 seconds The comparison between measurements (AB 209 cm and AC 308 cm)flight withwas the ones autonomous flight, at 20 mground-truth height and with a 70° camera angle. Furthermore, a manual obtained the final denseflight pointheight cloudduring (AB 210 1 cm and AC maintaining 309 ± 1 cm),the showed, alsofrom performed varying the±mission, while camera again, angle only constant. Thisbetween flight aimed to document vehicle portion, thus complementing data 1 cm discrepancy them, provingthe that bothobstructed ground and aerial imagery can be the combined to acquisition task. obtain a precise road traffic accident scenario reconstitution using the proposed methodology. Figure 8a presents this simulated crash scenario and the six points (A–F) marked on the ground for precision assessment. Figure 8b presents the final 3D model obtained through photogrammetric 5.4. Scenario 4: Presence of Buildings processing of the UAV-based imagery acquired by both the manual and automatic flights. This fourth scenario simulates a road traffic accident in which at least one of the involved vehicles is fairly close to a building, thus having one of its sides obstructed to autonomous aerial imagery acquisition. The aim is to depict a common situation, especially in urban contexts. Regarding UAV imagery, 40 images were obtained with an approximately 4 min and 30 s autonomous flight, at 20 m height and with a 70◦ camera angle. Furthermore, a manual flight was also performed varying flight height during the mission, while maintaining the camera angle constant. This flight aimed to document the vehicle obstructed portion, thus complementing the data acquisition task. Int. J. Environ. Res. Public Health 2020, 17, 1868 14 of 24 Figure 8a presents this simulated crash scenario and the six points (A–F) marked on the ground for precision assessment. Figure 8b presents the final 3D model obtained through photogrammetric processing of the UAV-based imagery acquired by both the manual and automatic flights. Int. J. Environ. Res. Public Health 2020, 17 14 of 24 Figure 8. Road traffic accident simulated scenario near buildings: (a) aerial overview of the scene Figure 8. Road traffic accident simulated scenario near buildings: (a) aerial overview of the scene along with the six control points (A–F) marked for precision assessment; (b) 3D model reconstituting along with the six control points (A–F) marked for precision assessment; (b) 3D model reconstituting the crash scene scene from from the the photogrammetric photogrammetric processing green dots dots represent represent the the the crash processing of of the the aerial aerial imagery, imagery, green camera positions. positions. camera Table77compares comparesthe the ground-truth measurements the ones the dense point Table ground-truth measurements withwith the ones takentaken from from the dense point cloud. cloud. A 1 cm maximum potential error in DE demonstrates the feasibility of this approach, proving A 1 cm maximum potential error in DE demonstrates the feasibility of this approach, proving that that imagery autonomous and manual may be combined to obtain a precise road imagery fromfrom both both autonomous and manual flightsflights may be combined to obtain a precise road traffic traffic accident scenario reconstitution using the proposed methodology. accident scenario reconstitution using the proposed methodology. Table7.7.Comparison Comparison between the ground-truth measurements withobtained those obtained from the Table between the ground-truth measurements with those from the generated generated cloud where is a building partially obstructing point cloudpoint where a building partially is obstructing a portion ofaaportion vehicle.of a vehicle. Distance Distance AB BC DE DF AB BC DE DF Tape (cm) Tape (cm) 459 459 251 251 268 268 669 669 Point Point Cloud (cm)(cm) Cloud 459 ±459 1 ±1 251 ±251 1 ±1 269 ±269 1 ±1 669 ±669 1 ±1 5.5. Scenario 5: Light Light Conditions Conditions obtaining reliable and accurate accurate Besides optimised flight parameters, imagery quality is key to obtaining traffic accident accident scenarios, scenarios, using using the the proposed proposed methodology. methodology. While While physical physical reconstitutions of road traffic obstructions to aerial imagery can be managed managed by using using complementary complementary acquisition equipment and central role role regarding regarding imagery imagery quality. quality. Insufficient light techniques, light conditions can play aa central accident context require artificial light sources to apply apply the proposed proposed conditions in a road traffic accident methodology,asasititis is highly dependent on the imagery quality to deliver complete and accurate methodology, highly dependent on the imagery quality to deliver complete and accurate digital digital outcomes. A lighting kitPhantom for DJI Phantom 4 UAVs was selected (Lume Cube, outcomes. A lighting kit for DJI 4 UAVs was selected (Lume Cube, Seattle, WA, Seattle, USA). ItWA, was USA). It was composed by two Light Diode (LED) lights—each a 3.81 cmincube, composed by two Light Emitting DiodeEmitting (LED) lights—each a 3.81 cm cube, weighting 28.35weighting g (battery ◦ beam in 28.35 g supplying (battery included), supplying at 1angle m, with a 60° beam angle andangle—and an adjustable included), 750 LUX at 1 m, with750 a 60LUX and an adjustable position all position angle—and all the necessary parts it inillumination, the UAV. As ground-based the necessary parts to assemble it in the UAV. As to for assemble ground-based fourfor Aputure Amaran illumination, lightsShenzhen, (AputureChina) Imaging Co. Ltd., AL-H198 LED four lightsAputure (AputureAmaran ImagingAL-H198 IndustriesLED Co. Ltd., wereIndustries selected. Each unit Shenzhen,23.4 China) selected. unit×measures 23.4 × 9.1 14 cm (5.5V–9.5V (length × width × height), measures × 9.1were × 14 cm (lengthEach × width height), powered by×battery DC), weighs 325 powered battery (5.5V–9.5V 325 g,including supplies 920 LUX atlight 1 m,intensity with a 60° beam angle, g, suppliesby920 LUX at 1 m, withDC), a 60◦weighs beam angle, adjustable and adjustable including adjustable light intensity and adjustable colour temperature. colour temperature. simulated road road traffic traffic accident accident scenario scenario under under adverse adverse illumination illumination conditions conditions was was The first simulated intended to visually assess acquired aerial imagery. The simulation simulation took took place in a suburban area, without direct direct light light from from nearby nearby public public illumination illumination and and involving involving two two vehicles. vehicles. Relying only on without portable illumination illuminationcoupled coupledtotothe theUAVs, UAVs,images imageswere weretaken taken the same point different heights portable inin the same point at at different heights (5 (5 m, 7.5 m, 10 m, 15 m, 20 m, and 25 m) over the scene, as shown in Figure 9. An image was acquired in a manual flight, with the UAVs RGB camera in a nadiral position. Int. J. Environ. Res. Public Health 2020, 17, 1868 15 of 24 m, 7.5 m, 10 m, 15 m, 20 m, and 25 m) over the scene, as shown in Figure 9. An image was acquired in a manual flight, with the UAVs RGB camera in a nadiral position. Int. Environ. Res. Public Health 2020, Int. J. J. Environ. Res. Public Health 2020, 1717 1515 ofof 2424 Figure9.9.Aerial Aerialimages imagesacquired acquiredwithout withoutpublic publicillumination illuminationsupport supportatatdifferent differentheights, heights,with with Figure Figure 9. Aerial images acquired without public illumination support at different heights, with portable portableillumination illuminationcoupled coupledtotothe theunmanned unmannedaerial aerialvehicle vehicle(UAV). (UAV). portable illumination coupled to the unmanned aerial vehicle (UAV). Portable illumination coupled theUAVs UAVs enough document road trafficaccidents accidents Portable illumination coupled totothe isisenough totodocument road traffic Portable illumination coupled to the UAVs is enough to document road traffic accidents scenarios scenarios withhigh high visual quality, under adverseillumination illumination conditions, heights equaloror inferior scenarios visual quality, under adverse conditions, heights inferior with high with visual quality, under adverse illumination conditions, to heightstoto equal or equal inferior to 15 m. to 15 m. Flights over that reference height cannot be dismissed, mainly because obstacles may to 15 m. Flights over that reference height cannot be dismissed, mainly because obstacles may bebe Flights over that reference height cannot be dismissed, mainly because obstacles may be present in the present in the scene. However, images visual quality decreases as flights height increases (Figure presentHowever, in the scene. However, images decreases visual quality decreases asincreases flights height increases (Figure 9).9). scene. images visual quality as flights height (Figure 9). Thesecond secondsimulation simulationunder under adverse illumination conditions took place parking lot,where where The adverse illumination conditions took place inina aparking lot, The second simulation under adverse illumination conditions took place in a parking lot, lighting was provided exclusively by public illumination fairly close to the area of interest. Regarding lighting was provided exclusively by public illumination fairly close to the area of interest. Regarding where lighting was provided exclusively by public illumination fairly close to the area of interest. aerialimagery, imagery, fourDG DGautonomous autonomous flightswere werecarried carried outwith with (Figure 10band andd)d)and andwithout without aerial flights out 10b Regarding aerialfour imagery, four DG autonomous flights were carried out(Figure with (Figure 10b,d) and without (Figure10a and c)the theportable portable illumination coupled the UAV: two mheight height (Figure10a 10aand and (Figure10a and illumination coupled totothe UAV: atat2020m (Figure (Figure 10a,c) thec)portable illumination coupled to the UAV: two at 20two m height (Figure 10a,b)—25 images b)—25 images were acquired in approximately 4 minutes and 30 seconds—and another two at 30 b)—25 images were acquired in approximately 4 minutes and 30 seconds—and another two at 30 mm were acquired in approximately 4 min and 30 s—and another two at 30 m height (Figure 10c,d)—17 height(Figure (Figure10c 10c andd)—17 d)—17images images acquired approximately minute and seconds. All flights ◦ camera height acquired minute and 4040seconds. flights images acquired inand approximately 1 min and 40in s.inapproximately All flights had a1 170 angle and theAll portable had a 70° camera angle and the portable illumination coupled to the UAV was manually adjusted had a 70° camera angle and the portable illumination coupled to the UAV was manually adjusted toto illumination coupled to the UAV was manually adjusted to this angle. Lower flight heights were not this angle. Lower flight heights were not considered due to the presence of trees and aerial cables this angle. Lower heightsof were not considered dueintothe thescene presence of trees and aerial cables inin considered due to flight the presence trees and aerial cables surroundings. thescene scenesurroundings. surroundings. the Figure10. 10.Orthophoto Orthophotomosaics mosaics generatedfrom from aerialimagery imagery acquiredininananadverse adverse lightcontext: context: Figure Figure 10. Orthophoto mosaics generated generated from aerial aerial imagery acquired acquired in an adverse light light context: (a) and (c) only supported by public illumination; (b) and (d) public illumination combined with the (a) and and (c) (c) only only supported supported by by public public illumination; illumination; (b) (b) and and (d) (d) public public illumination illumination combined combined with with the the (a) portableone onecoupled coupledtotothe theUAV; UAV;(a)(a)and and(b) (b)atat2020mmheight; height;and and(c)(c)and and(d) (d)atat3030mmheight. height. portable portable one coupled to the UAV; (a) and (b) at 20 m height; and (c) and (d) at 30 m height. Allthe thepresented presentedoutputs outputsdisplayed displayedquality qualityreconstitutions reconstitutionsof the scenario.Although Althoughpublic public All presented outputs displayed quality reconstitutions ofofthe the scenario. scenario. illumination was sufficient in this scenario, portable illumination coupled to UAVs can also be used illumination was was sufficient sufficientin inthis thisscenario, scenario,portable portableillumination illuminationcoupled coupled UAVs also used illumination toto UAVs cancan also be be used as complement, improving digitaloutputs outputs lightintensity/colour intensity/colour andcompensating compensating possible a acomplement, improving light and possible aasas complement, improving digitaldigital outputs light intensity/colour and compensating possible insufficient insufficient publicillumination illumination (e.g., poles thatare are toomissing spacedout, out, missingorormalfunctioning). malfunctioning). insufficient public (e.g., poles that too spaced missing public illumination (e.g., poles that are too spaced out, or malfunctioning). thirdsimulation simulationunder underadverse adverseillumination illuminationconditions conditionswas wasset setinina asuburban suburbanarea, area,without without AAthird publiclighting lightingsupport. support.Two Twovehicles—one vehicles—oneofofdarker darkerand andone oneofofbrighter brightercolour—were colour—wereused usedand andthe the public accident scene was documented using: (i) only ground-based portable illumination (four units); (ii) accident scene was documented using: (i) only ground-based portable illumination (four units); (ii) onlyportable portableillumination illuminationcoupled coupledtotothe theUAV; UAV;and and(iii) (iii)a acombination combinationbetween betweenground-based ground-basedand and only UAV-coupledillumination. illumination. UAV-coupled Int. J. Environ. Res. Public Health 2020, 17, 1868 16 of 24 A third simulation under adverse illumination conditions was set in a suburban area, without public lighting support. Two vehicles—one of darker and one of brighter colour—were used and the accident scene was documented using: (i) only ground-based portable illumination (four units); (ii) only portable illumination coupled to the UAV; and (iii) a combination between ground-based and Int. J. Environ. Res. Public Health 2020, 17 16 of 24 UAV-coupled illumination. Regarding aerial imagery, several DG autonomous flights were carried out. Each flight had a 70◦ Regarding aerial imagery, several DG autonomous flights were carried out. Each flight had a 70° camera angle and the portable illumination was manually adjusted to this angle. The 25 m height camera angle and the portable illumination was manually adjusted to this angle. The 25 m height flight—28 images acquired in approximately 5 min and 30 s—was considered for this scenario for two flight—28 images acquired in approximately 5 minutes and 30 seconds—was considered for this reasons: it is borderline regarding aerial imagery quality and this flight height allows for the presence scenario for two reasons: it is borderline regarding aerial imagery quality and this flight height allows of most type of obstacles. Figure 11 presents the final orthophoto mosaics focused on the vehicles for the presence of most type of obstacles. Figure 11 presents the final orthophoto mosaics focused involved in the simulation, for each aforementioned illumination setup. on the vehicles involved in the simulation, for each aforementioned illumination setup. Figure 11. Orthophoto mosaics generated from aerial imagery acquired in an adverse light context: Figure 11. Orthophoto mosaics generated from aerial imagery acquired in an adverse light context: (a) only ground-based illumination; (b) portable illumination coupled to the unmanned aerial vehicle (a) only ground-based illumination; (b) portable illumination coupled to the unmanned aerial vehicle (UAV); and (c) combination of both ground and UAV-coupled illumination. Yellow circles mark the (UAV); and (c) combination of both ground and UAV-coupled illumination. Yellow circles mark the location of the four ground-based illumination units. location of the four ground-based illumination units. The best best visual visual result result seems seems to to be be the the one one only only using The using portable portable illumination illumination coupled coupled to to the the UAV UAV (Figure 11b), as it seems to reduce shadow effects present in the combination between both portable (Figure 11b), as it seems to reduce shadow effects present in the combination between both portable illuminations (Figure 11c). Moreover, has simpler simpler logistics logistics for for police police forces forces when when compared compared illuminations (Figure 11c). Moreover, it it also also has with ground-based illumination that requires setting up units in the crash scene. with ground-based illumination that requires setting up units in the crash scene. 6. Experimental Tests: Real Scenarios traffic accidents accidents scenarios, scenarios, addressed in cooperation cooperation with This section presents some real road traffic both PSP and GNR, in which the proposed methodology is compared with the current reconstitution method used by police forces. 6.1. Urban Context Within Within the the scope of a close collaboration with PSP of Vila Real, our research group accompanied PSP officers to some real real road road traffic traffic accidents accidents that that happened happened within within Vila Vila Real Real urban urban context. context. Both teams worked roughly simultaneously to document crash scenes. Any road traffic accident scenario was disregarded disregarded if if any any of the involved parties did not concede authorization to disclose the acquired images. most meaningful documented occurrences are presented. using theirusing own approaches: images. Only Onlythethe most meaningful documented occurrences are presented. their own the current reconstitution method and the proposed methodology, respectively. PSP sketches produced approaches: the current reconstitution method and the proposed methodology, respectively. PSP for the presented occurrences were provided for both and validation purposes. sketches produced for the presented occurrences werecomparison provided for both comparison and validation The first occurrence regards a daytime urban road traffic accident—one-way street, with two purposes. trafficThe lanes and parking inregards both sides of the street—that took placeaccident—one-way on 10 April 2017, directly first occurrence a daytime urban road traffic street, involving with two two vehicles. PSP parking was called the accident and proceeded documenting it using its directly current traffic lanes and in toboth sides of scene the street—that tookbyplace on 10 April 2017, methodology 12c). sketch in Microsoft Visio scene (Microsoft Corporation, Redmond, USA) was involving two(Figure vehicles. PSPAwas called to the accident and proceeded by documenting it using produced using the data its currentafterwards methodology (Figure 12c).acquired. A sketch in Microsoft Visio (Microsoft Corporation, Redmond, USA) was produced afterwards using the data acquired. As for the proposed methodology, a DG autonomous flight was carried out roughly at the same time that PSP officers were doing their investigation. The 20 m height flight—due to the presence of public illumination poles in the surroundings of the scene—50 images were acquired in approximately two minutes, with a 70° camera angle and 80% overlap between images. Figure 12a Int. J. Environ. Res. Public Health 2020, 17 17 of 24 vehicles passing by are depicted as ghost-like elements, but allow for comprehending traffic flow during the period of time lapsed between the road traffic accident and the end of PSP’s documenting Int. J. Environ. Res. Public Health 2020, 17, 1868 of 24 process. Figure 12b presents the processed orthophoto mosaic without the ghost-like elements17(e.g., vehicles and people on the sidewalk), allowing the viewer to focus on the crash scene elements only. Figure 12. Proposed method applied to a road traffic accident in urban context: (a) orthophoto mosaic Figure 12. Proposed method applied to a road traffic accident in urban context: (a) orthophoto generated from the aerial imagery photogrammetric processing; (b) orthophoto mosaic without mosaic generated from the aerial imagery photogrammetric processing; (b) orthophoto mosaic without ghost-like elements; elements; (c) (c)Police Policeofficers officerstaking takingmeasurements measurementswith witha atape tape and the UAV getting ready ghost-like and the UAV getting ready to to take-off; (d) aerial perspective of the scene, yellow ellipses identify the vehicles involved in the take-off; (d) aerial perspective of the scene, yellow ellipses identify the vehicles involved in the accident; accident; (e) PSP’s sketch; and (f)between overlap PSP’s between PSP’s sketch and the generated orthophoto (e) PSP’s sketch; and (f) overlap sketch and the generated orthophoto mosaic. mosaic. Figure 12f proposed presents an overlap between sketch (Figure 12e)carried and the processed orthophoto As for the methodology, a DG PSP’s autonomous flight was out roughly at the same mosaic 12b), both the their sameinvestigation. scale. There are regarding vehicles’ time that(Figure PSP officers were in doing Thesome 20 m minor height issues flight—due to thethe presence of dimensions. However, considering that PSP officers use existing models from Microsoft Visio to public illumination poles in the surroundings of the scene—50 images were acquired in approximately ◦ develop their work, scenario bothand the 80% sketch and the orthophoto mosaic are very similar. As two minutes, with ain70this camera angle overlap between images. Figure 12a presents thea road traffic accident scenario reconstitution, the proposed approach provides a more complete set of orthophoto mosaic, reconstituting this road traffic accident scenario. Some debris and possible elements outcomes, enabling different perspectives to be analysed (Figure 12d), zoom in a certain area, assess of interest can be clearly distinguished lying on the road. In the open traffic lane, vehicles passing by the depicted overall accident scene environmental and physical contexts, precise of debris other are as ghost-like elements, but allow for comprehending trafficlocation flow during the and period of elements of interest and capability to measure distances between any two points within the scene. time lapsed between the road traffic accident and the end of PSP’s documenting process. Figure 12b Moreover, accidentorthophoto scene is digitally preserved analysis when necessary. Thisand relatively presents thethe processed mosaic without the for ghost-like elements (e.g., vehicles people simple urban road traffic accident only down one traffic lane for approximately 25 minutes on the sidewalk), allowing the viewer to closed focus on the crash scene elements only. (on-site documenting process), which meant PSP’s that traffic still12e) flowand in the trafficorthophoto lane. With Figure 12f presents an overlap between sketchcould (Figure theother processed the proposed approach, documenting crashThere scene are tooksome only minor two minutes, with five more minutes mosaic (Figure 12b), both in the samethe scale. issues regarding the vehicles’ spent in assembling the UAV and setting up the flight plan. dimensions. However, considering that PSP officers use existing models from Microsoft Visio to A second occurrence also a daytime urban traffic accident, a busy develop their work, in thisregards scenario both the sketch androad the orthophoto mosaic are intersection, very similar.that As place on accident May 11, 2017, involving two vehicles. PSP was also called to the traffic accident scene atook road traffic scenario reconstitution, the proposed approach provides a more complete andofproceeded documenting it perspectives using its current produced sketchinisa presented in set outcomes,by enabling different to bemethods. analysedThe (Figure 12d), zoom certain area, Figure 13a. Two traffic signs—a stop (PFA) and a wrong way (PFB) sign—and the collision point (CP) assess the overall accident scene environmental and physical contexts, precise location of debris and are considered as interest reference points. PSP to officers tookdistances around 35 minutes complete on-site other elements of and capability measure between anytotwo points the within the documenting process. As for the proposed methodology, a DG autonomous flight was carried out. scene. Moreover, the accident scene is digitally preserved for analysis when necessary. This relatively The 20urban m height flight—due to only the presence of public illumination poles and buildings in the simple road traffic accident closed down one traffic lane for approximately 25 min (on-site surroundings—acquired 57 images in approximately 2 minutes, with a 70° camera angle and documenting process), which meant that traffic could still flow in the other traffic lane. With80% the imagery overlap. Figure 13b presents the produced mosaicwith of this accident proposed approach, documenting the crash scene took orthophoto only two minutes, fiveroad moretraffic minutes spent scenario. The the three points considered in assembling UAVreference and setting up the flight plan.in PSP’s sketch (Figure 13a) are highlighted. Furthermore, debris and other possible elements of interest be clearly distinguished lying onthat the A second occurrence regards also a daytime urban roadcan traffic accident, a busy intersection, road, near the collision point (PC). took place on May 11, 2017, involving two vehicles. PSP was also called to the traffic accident scene and proceeded by documenting it using its current methods. The produced sketch is presented in Figure 13a. Two traffic signs—a stop (PFA) and a wrong way (PFB) sign—and the collision point (CP) are considered as reference points. PSP officers took around 35 min to complete the on-site documenting process. As for the proposed methodology, a DG autonomous flight was carried out. The 20 m height flight—due to the presence of public illumination poles and buildings in the surroundings—acquired 57 images in approximately 2 min, with a 70◦ camera angle and 80% imagery overlap. Figure 13b Int. J. Environ. Res. Public Health 2020, 17, 1868 18 of 24 presents the produced orthophoto mosaic of this road traffic accident scenario. The three reference points considered in PSP’s sketch (Figure 13a) are highlighted. Furthermore, debris and other possible elements of interest can be clearly distinguished lying on the road, near the collision point (PC). Int. J. Environ. Res. Public Health 2020, 17 18 of 24 Figure 13. Sketch from Public Safety Police, based on data documented at the traffic accident scene Figure 13. Sketch from Public Safety Police, based on data documented at the traffic accident scene with with two vehicles involved (a). Orthophoto mosaic generated from the photogrammetric processing two vehicles involved mosaicpoints generated from in thePSP’s photogrammetric processing of aerial imagery(a). (b),Orthophoto the three reference considered sketch are highlighted usingof aerial imageryyellow (b), the three reference points considered in PSP’s sketch are point highlighted usingbetween yellow circles: circles: two traffic signs (yellow enclosed “×”) and the collision (PC). Overlap the accident sketch and the orthophoto mosaic, ellipses mark the discrepancies regarding two traffic signs (yellow enclosed “×”) and the white collision point (PC). Overlap between thetheaccident position of both vehicles (c). sketch and the orthophoto mosaic, white ellipses mark the discrepancies regarding the position of both vehicles (c). Figure 13c presents an overlap between PSP’s sketch (Figure 13a) and the processed orthophoto mosaic (Figure 13b), both in the same scale. The main noticeable discrepancy regards the position of Figure 13c presents an overlap between PSP’s sketch (Figure 13a) and the processed orthophoto the vehicles. In a more complex road traffic accident scenario, the proposed approach would provide mosaic a(Figure 13b), both in the same The main noticeable discrepancy regards the position of more precise reconstitution, withscale. a comprehensive context allowing for different perspectives, the vehicles. In a more complex road traffic accident scenario, the proposed approach would provide a precise location of the debris, other elements of interest and the capability of measuring distances between any two points within the scene. Again, the crash scene is fully preserved for both present more precise reconstitution, with a comprehensive context allowing for different perspectives, precise future analysis. Thiselements urban road accident took approximately 35 minutes to be locationand of the debris, other oftraffic interest and scenario the capability of measuring distances between documented by PSP, with some impact in the usual traffic flow. With the proposed approach, any two points within the scene. Again, the crash scene is fully preserved for both present and future documenting the entire crash scene took only 2 minutes, with 5 more minutes spent in assembling analysis. This urban road traffic accident scenario took approximately 35 min to be documented by the UAV and setting up a proper flight plan. PSP, with some impact in the usual traffic flow. With the proposed approach, documenting the entire 6.2. Motorway Scenario crash scene took only 2 min, with 5 more minutes spent in assembling the UAV and setting up a proper flight plan. A motorway in the outskirts of Vila Real (IP4) is the last road traffic accident scenario, in real environment, presented in this work. It consisted on a simulated scene between two vehicles in an 6.2. Motorway Scenario IP4 stretch with two lanes in one direction and one lane in another. The scenario was created to be particularly complex due to the overall physical context and to the scattered debris field, formed by A motorway in the outskirts of Vila Real (IP4) is the last road traffic accident scenario, in real objects of reduced dimensions (when compared to those of a vehicle), by skid marks and by an oil environment, in one thisofwork. It consisted a simulated sceneshown between two vehicles spill, putpresented right next to the vehicles involvedon (different perspectives in Figure 14a–d). in an IP4 stretch with two lanes in one direction and one lane in another. The scenario was created GNR was responsible both for staging the road traffic accident and by meticulously documenting it to be using complex its current due method: a measuring wheel (Figure 14e), aand measuring and a hand drawn sketch particularly to the overall physical context to the tape scattered debris field, formed by 14f). Officers took approximately 3 hourstotothose document accident sketch was objects (Figure of reduced dimensions (when compared of a the vehicle), byscene. skid The marks and by an oil then used to produce two sketches in Microsoft Visio: one non-scaled—that took two hours to be spill, put right next to one of the vehicles involved (different perspectives shown in Figure 14a–d). done—and a scaled one (Figure14g), that took 16 hours to be completed. GNR was responsible both for staging the road traffic accident and by meticulously documenting it using its current method: a measuring wheel (Figure 14e), a measuring tape and a hand drawn sketch (Figure 14f). Officers took approximately 3 h to document the accident scene. The sketch was then used to produce two sketches in Microsoft Visio: one non-scaled—that took two hours to be done—and a scaled one (Figure 14g), that took 16 h to be completed. Int. J. Environ. Res. Public Health 2020, 17 19 of 24 Int. J. Environ. Res. Public Health 2020, 17, 1868 19 of 24 Int. J. Environ. Res. Public Health 2020, 17 19 of 24 Figure 14. IP4 motorway road traffic accident scenario: different perspectives from the involved Figure 14. IP4 motorway road traffic accident scenario: different perspectives from the involved vehicles debris (a,road b, c, and d); National officers documentingfrom the accident Figure 14. IP4 and motorway traffic accidentRepublican scenario: Guard different perspectives the involved vehiclesscene and debris (a, b, c,handmade and d); National Republican Guard officersbydocumenting the accident scene (e); the on-site sketch (f); and scaled sketch produced GNR in office (g). vehicles and debris (a, b, c, and d); National Republican Guard officers documenting the accident (e); the on-site handmade sketch (f); and scaled sketch produced by GNR in office (g). scene (e); the on-site handmade sketch (f); and scaled sketch produced by GNR in office (g). As for the proposed methodology, a DG autonomous flight was carried out. The 25 m height encompassed 67 × 40 m area,a acquiring 135 imagesflight in approximately minutes Asflight for the proposed amethodology, DG autonomous was carried5 out. Theand 25 30 m height As seconds, for the with proposed methodology, a80% DGimagery autonomous flight was carried out.a The 25 m height◦ a 70° camera angle and overlap. In only a few minutes, substantial flight encompassed a 67 × 40 m area, acquiring 135 images in approximately 5 min and 30 s, with a 70 flight encompassed a 67 40 mscene area, acquiring 135quickly images in approximately 5 minutes the× crash dimensions—was documented. Furthermore, the high- and 30 camera area—considering angle and 80% imagery overlap. In only a few minutes, a substantial area—considering the seconds,level with a 70°achieved cameracan angle and 80% imagery In the only a afew minutes, substantial of detail be seen in Figure 15. Debrisoverlap. scattered on road, car wheel in theaditch, crash scene dimensions—was quickly documented. Furthermore, the high-level of detail achieved can skid marks and cones circumscribing the accident scene are clearly identifiable. area—considering thetraffic crash scene dimensions—was quickly documented. Furthermore, the highbe seen in Figure 15. Debris scattered on the road, a car wheel in the ditch, skid marks and traffic cones level of detail achieved can be seen in Figure 15. Debris scattered on the road, a car wheel in the ditch, circumscribing the accident scene are clearly identifiable. skid marks and traffic cones circumscribing the accident scene are clearly identifiable. Figure 15. Part of the orthophoto mosaic of the IP4 motorway road traffic accident scenario (a). Both vehicles and other elements of interest (white ellipses): debris (c), skid marks (d), a tire (b). Measurement lines are also identified (A to H). Figure 15. Part of the orthophoto mosaic of the IP4 motorway road traffic accident scenario (a). Both Figure 15. Part of the orthophoto mosaic of the IP4 motorway road traffic accident scenario (a). vehicles and other elements of interest (white ellipses): debris (c), skid marks (d), a tire (b). Both vehicles and other elements of interest (white ellipses): debris (c), skid marks (d), a tire (b). Measurement lines are also identified (A to H). Measurement lines are also identified (A to H). Int. J. Environ. Res. Public Health 2020, 17, 1868 20 of 24 Int. J. Environ. Res. Public Health 2020, 17 20 of 24 Figure 16 presents an overlap between the scale sketch (Figure 14g) and the processed orthophoto Figure 16 presents an overlap between the scale sketch (Figure 14g) and the processed mosaic (Figure 15a), both in the same scale. Notwithstanding the GNR officers experience and orthophoto mosaic (Figure 15a), both in the same scale. Notwithstanding the GNR officers experience the painstaking documenting process carried out at the accident scene, there are some important and the painstaking documenting process carried out at the accident scene, there are some important discrepancies between the sketch and the orthophoto mosaic, independently on how they are aligned. discrepancies between the sketch and the orthophoto mosaic, independently on how they are aligned. Both images were put to the same scale and lined up considering motorway overpasses and the Both images were put to the same scale and lined up considering motorway overpasses and the motorway guard rails next to the vehicles. Regarding the designated area of interest, debris positions motorway guard rails next to the vehicles. Regarding the designated area of interest, debris positions are not correct, as are also the vehicles positions and skid marks. In a broader context, both the are not correct, as are also the vehicles positions and skid marks. In a broader context, both the motorway profile and the respective reference elements do not match. Again, the proposed approach motorway profile and the respective reference elements do not match. Again, the proposed approach provides a more precise reconstitution, with a comprehensive context and debris precise location. provides a more precise reconstitution, with a comprehensive context and debris precise location. Figure 16. Overlap between National Republican Guard’s (GNR’s) sketch and the generated Figure 16. Overlap between National Republican Guard’s (GNR’s) sketch and the generated orthophoto mosaic. orthophoto mosaic. To further validate validate the the proposed proposed methods methods capability To further capability to to preserve preserve such such aa complex complex road road traffic traffic accident scenario for both present and future analysis, different perspectives of the traffic accident accident scenario for both present and future analysis, different perspectives of the traffic accident scene are shown shownin inFigure Figure1717and andthat that which determines ability to measure distances between scene are which determines thethe ability to measure distances between any any two points within the scene. Based on GNR officers work in documenting the crash scene, Table two points within the scene. Based on GNR officers work in documenting the crash scene, Table 8 8presents presents a comparison between on-site measurements andobtained those obtained the proposed a comparison between on-site measurements and those from thefrom proposed approach. approach. eachdistance measured distance is represented Figure 15a for purposes. clarification purposes. Moreover, Moreover, each measured is represented in Figure 15ainfor clarification Considering Considering tolerances, thepossible maximum possible error of 6 cm in measurement (4% error), taken tolerances, the maximum error is of 6 cm in is measurement F (4% error), Ftaken between the between thegreen front vehicle of the green vehicle front of the and the road and line.the road line. This motorway road traffic accident scenario took approximately 16 office hours to be represented in an official sketch. With the proposed approach, documenting the entire crash scene took approximately 5 min and 30 s (with 5 min more spent in assembling the UAV and setting up a proper flight plan) and approximately 30 min in the office to obtain the digital outputs. In addition to being faster, the proposed approach also enables precise measurements to be taken between two points in the crash scene, as well as different perspectives of the accident with a broader context. The digital outputs of the road traffic accident scenario reconstitution are made available for future analysis, if necessary. Figure 17. Different 3D perspectives from the road traffic accident scenario in a motorway (IP4). To further validate the proposed methods capability to preserve such a complex road traffic accident scenario for both present and future analysis, different perspectives of the traffic accident scene are shown in Figure 17 and that which determines the ability to measure distances between any two points within the scene. Based on GNR officers work in documenting the crash scene, Table 8 presents a comparison between on-site measurements and those obtained from the proposed approach. Moreover, each measured Int. J. Environ. Res. Public Health 2020, 17, 1868 distance is represented in Figure 15a for clarification purposes. 21 of 24 Considering tolerances, the maximum possible error is of 6 cm in measurement F (4% error), taken between the front of the green vehicle and the road line. Figure 17. Different 3D perspectives from the road traffic accident scenario in a motorway (IP4). Figure 17. Different 3D perspectives from the road traffic accident scenario in a motorway (IP4). Table 8. Comparison between measurements taken by National Republican Guard (GNR) officers with those obtained from the digital products generated by the proposed approach, for the motorway (IP4) road traffic accident. Errors associated with the proposed methodology measurements were calculated using Pix4D. Distance Line Tape (cm) Point Cloud (cm) A B C D E F G H 163 163 163 260 252 143 300 70 166 ± 1 163 ± 3 162 ± 3 260 ± 2 254 ± 4 137 ± 4 300 ± 2 71 ± 1 7. Conclusions Clearly, UASs allow for collecting more data with higher precision when compared to the current approaches for documenting road traffic accident scenes. However, in certain conditions, some field data retrieval may not be possible due to the presence of obstacles in the scenario—both in urban and rural contexts—such as trees, cables or buildings. Moreover, UAV-based photogrammetric methods can deal with adverse light conditions which remain scarcely addressed, leaving law enforcement authorities with no other option than to resort to the traditional approaches or others based on expensive and/or laboriously demanding technologies, whenever a road occurrence requiring their intervention takes place, for example, at night or on cloudy days. To tackle these issues, this study proposes and evaluates a low cost UAV-based photogrammetric methodology covering a wide range of scenarios, including: (i) obstacle-free scenes using only UAS as surveying instrument; (ii) scenes featured by road traffic accidents with partial inaccessibility for UAS usage, thus requiring complementary acquisition approaches; (iii) simulations in adverse light contexts (e.g., at night or with bad visibility), requiring the use of artificial light sources; and (iv) real road traffic accidents. The several tests performed in this study and the literature review enabled us to obtain the benefits and limitations regarding the available approaches to address the digital reconstitution road traffic accident scenes, which are presented in Table 9. Despite the number of valid traffic accident research works involving the use of UAV-based surveying, most of them were developed considering ideal scenarios, either on urban or rural environments. To address this gap, a methodology consisting of a full pipeline to digitally document road traffic accidents using UAVs, ground cameras, artificial light sources and photogrammetry was Int. J. Environ. Res. Public Health 2020, 17, 1868 22 of 24 proposed in this technical note. A preliminary set of experiments were carried out to determine default flight parameters on ideal scenarios, which can be fine-tuned according with each set of conditions characterising a road traffic accident occurrence. Afterwards, to demonstrate the usefulness of UAVs on traffic accidents investigation, a set of experiments was performed, more specifically: (i) considering areas populated with typical obstacles found in urban and rural settings, at different densities; and (ii) under unfavourable illumination conditions. Furthermore, practical situations were addressed with the Portuguese road traffic authorities, wherein traditional approach and proposed methodology are compared in terms of efficiency and accuracy. Table 9. Comparison between the available approaches to address road traffic accident scenes reconstitution: pros, cons and possible outputs. Approach Pros Cons Outputs Measuring Tape or Wheel 3Ease-of-use 3Low cost ×Time-consuming ×Prone to errors ×No visual data ×Data not georeferenced -Manual measurements Sketches & Photographs 3Ease-of-use 3Low cost ×Time-consuming ×Inaccurate ×Data not georeferenced -Measurements -Images -Forensic sketches Total Station 3Rigorous ×Time-consuming ×Skilled operator ×Not suitable for mobility ×Considers line-of-sight ×Expensive -Precise measurements (angles and distances) Laser Scanning 3Rigorous 33D Visualisation ×Time-consuming ×Loss of information due to its position ×Confined action range ×Expensive -Point clouds -Images GNSS Receivers 3Rigorous ×Time-consuming ×Prone to signal obstruction ×Expensive -Precise georeferenced points UAS 3Ease-of-use 3Large coverage 3Multiple outputs 33D Visualisation 3Cost-effective ×Data acquisition constrains ×Liable to collisions ×Requires training ×Rigidly Regulated -Point clouds -3D meshes -Orthophoto mosaics -Digital Surface Models Encouraging results were obtained from the assessment of the different experiments, which attest the feasibility of using UAVs for traffic accident investigation in many environments characterised by several conditions. Regarding extreme situations wherein traffic accident scenarios are under severe influence of obstructing obstacles, other alternative/complementary strategies to digitally document evidences can be used at the surveying stage, by using, for example consumer-grade digital cameras or smartphones. Moreover, in regards to the tests carried out in real situations, in simpler scenarios the results of both traditional approaches and the proposed methodology reached a fair level of similarity. However, in more complex cases, significant advantages of the latter were observed in relation to the former in terms of accuracy and operation times. The complete solution implementing the proposed methodology requires the integration of the following components: (1) aerial platform (UAS), endowed with artificial lighting to operate in the most varied illumination conditions; (2) several types of sensors, according to the specificity of the operation and; (3) a computer application—freely available or developed from scratch—to generate the required digital documentation products, for consistent and non-redundant storing, enabling a quick assessment and retrieval of reliable measurements. Forthcoming directions include: motivating the change on actual traffic road accidents documentation paradigm with promotion and training initiatives—that have already started to take place with local authorities—aiming to complement and, eventually, replace the traditional approach with a technological pipeline, pursuing not only a normalised digital documentation process but also the reduction of required times and endeavour Int. J. Environ. Res. Public Health 2020, 17, 1868 23 of 24 in current practices carried out in the field for surveying purposes. Another important issue to address in future is the need to have UAVs operating in a wider number of nonextreme—though unfavourable—weather situations, in which structural upgrades such as UAV shield adjustments for waterproofing and hardware improvements for fast compensation of windy conditions may be expected, eventually, under the manufacturer’s initiative. The improvement of the methodology’s pipeline by implementing features for parallel processing while the UAV is surveying road traffic accidents, suppressing the need for post-processing activities in the office should also be addressed. Real-time online access to reconstructed road traffic accident scenes is another relevant capability to be developed. Moreover, layers of computational intelligence and analytics are being designed to enhance the proposed methodology with decision support features that may assist authorities, urban planning professionals and municipal entities in the continuous improvement of the road traffic system. Author Contributions: Conceptualization, J.J.S.; data curation, L.P., J.S., J.V., J.H. and T.A.; formal analysis, L.P., J.S. and J.V.; funding acquisition, E.P., and J.J.S.; investigation, L.P., J.S., J.V., J.H. and T.A.; methodology, A.S., E.P. and J.J.S.; project administration, E.P. and J.J.S.; resources, J.S., E.P. and J.J.S.; software, L.P., J.V. and T.A.; supervision, E.P., and J.J.S.; validation, L.P., J.V, E.P. and J.J.S.; visualization, L.P., J.S. and E.P.; writing—original draft, L.P., J.S., J.V., T.A., and E.P.; writing—review and editing, A.S., and J.J.S. All authors have read and agreed to the published version of the manuscript. Funding: This work was partially financed by the European Regional Development Fund (ERDF) through the Operational Programme for Competitiveness and Internationalisation—COMPETE 2020 under the PORTUGAL 2020 Partnership Agreement, and through the Portuguese National Innovation Agency (ANI) as a part of project “REVEAL—Drones for supporting traffic accident evidence acquisition by Law Enforcement Agents” (Nº 33113). Financial support provided by the FCT—Portuguese Foundation for Science and Technology (SFRH/BD/139702/2018) to Luís Pádua. Acknowledgments: The authors would like to acknowledge all the law enforcement agents that assisted with this study. Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. ETSC. Briefing: 5th EU Road Safety Action, 2020–2030; European Transport Safety Council, European Union: Brussels, Belgium, 2018; p. 47. WHO. Global Status Report on Road Safety 2015; World Health Organization: Geneva, Switzerland, 2015; p. 340. ITF. Road Safety Annual Report 2017; OECD Publishing: Paris, France, 2018; p. 586. Wijnen, W.; Stipdonk, H. Social costs of road crashes: An international analysis. Accid. Anal. Prev. 2016, 94, 97–106. [CrossRef] [PubMed] Rolison, J.J.; Regev, S.; Moutari, S.; Feeney, A. What are the factors that contribute to road accidents? An assessment of law enforcement views, ordinary drivers’ opinions, and road accident records. Accid. Anal. Prev. 2018, 115, 11–24. [CrossRef] [PubMed] Wegman, F. The future of road safety: A worldwide perspective. IATSS Res. 2017, 40, 66–71. [CrossRef] Elvik, R.; Høye, A.; Vaa, T.; Sørensen, M. Driver Training and Regulation of Professional Drivers’, The Handbook of Road Safety Measures; Emerald Group Publishing Limited: Bingley, UK, 2009. WHO. Save LIVES—A Road Safety Technical Package; World Health Organization: Geneva, Switzerland, 2017; p. 60. Ma, C.; Yang, D.; Zhou, J.; Feng, Z.; Yuan, Q. Risk Riding Behaviors of Urban E-Bikes: A Literature Review. Int J. Environ. Res. Public Health 2019, 16, 2308. [CrossRef] [PubMed] Ma, C.; Hao, W.; Xiang, W.; Yan, W. The Impact of Aggressive Driving Behavior on Driver-Injury Severity at Highway-Rail Grade Crossings Accidents. J. Adv. Transp. 2018, 2018, 1–10. [CrossRef] Wach, W. Structural reliability of road accidents reconstruction. Forensic Sci. Int. 2013, 228, 83–93. [CrossRef] [PubMed] Topolšek, D.; Herbaj, E.A.; Sternad, M. The Accuracy Analysis of Measurement Tools for Traffic Accident Investigation. J. Transp. Technol. 2014, 4, 84–92. [CrossRef] Int. J. Environ. Res. Public Health 2020, 17, 1868 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 24 of 24 Lowe, D.G. Object recognition from local scale-invariant features. In Proceedings of the Seventh IEEE International Conference on Computer Vision, Kerkyra, Greece, 20–27 September 1999; IEEE: Kerkyra, Greece, 1999; Volume 2, pp. 1150–1157. Dellaert, F.; Seitz, S.M.; Thorpe, C.E.; Thrun, S. Structure from motion without correspondence. In Proceedings of the Proceedings IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2000 (Cat. No.PR00662), Hilton Head Island, SC, USA, 15 June 2000; Volume 2, pp. 557–564. Arnold, E. Use of Photogrammetry as a Tool for Accident Investigation and Reconstruction: A Review of the Literature and State of the Practice; Virginia Transportation Research Council: Charlottesville, VA, USA, 2007. Cooner, S.A.; Balke, K.N. Use of Photogrammetry for Investigation of Traffic Incident Scenes; National Technical Information Service: Springfield, VA, USA, 2000. Zuraulis, V.; Levulytė, L.; Sokolovskij, E. Vehicle Speed Prediction from Yaw Marks Using Photogrammetry of Image of Traffic Accident Scene. Procedia Eng. 2016, 134, 89–94. [CrossRef] Fraser, C.; Cronk, S.; Hanley, H. Close-range photogrammetry in traffic incident management. In Proceedings of the XXI ISPRS Congress Commission V, WG V; The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences: Beijing, China, 2008; Volume XXXVII, pp. 125–128. Pádua, L.; Vanko, J.; Hruška, J.; Adão, T.; Sousa, J.J.; Peres, E.; Morais, R. UAS, sensors, and data processing in agroforestry: A review towards practical applications. Int. J. Remote. Sens. 2017, 38, 2349–2391. [CrossRef] Monadrone Accident Reconstruction & Evidence Gathering. Available online: http://www.monadrone.com/ applications/mission-summaries/accident-reconstruction-evidence-gathering/ (accessed on 5 October 2016). Pix4D. A New Protocol of CSI for the Royal Canadian Mounted Police. Available online: https: //assets.ctfassets.net/go54bjdzbrgi/2ufTRuEockU20mWCmosUmE/88ad2758f8c1042ab696c974627239e6/ RCMP_Pix4D_collision_crime_scene_investigation_protocol.pdf (accessed on 21 August 2018). Analist Group Traffic Accident Reconstruction. Available online: http://www.analistgroup.com/en/newsolutions-with-drones/solution-for-traffic-accident-reconstruction (accessed on 5 October 2016). senseFly. Setting the Scene—Using Drones for Traffic Accident Reconstruction and Analysis—Waypoint. Available online: http://waypoint.sensefly.com/setting-the-scene-using-drones-for-crash-reconstructionand-analysis/ (accessed on 5 October 2016). Su, S.; Liu, W.; Li, K.; Yang, G.; Feng, C.; Ming, J.; Liu, G.; Liu, S.; Yin, Z. Developing an unmanned aerial vehicle-based rapid mapping system for traffic accident investigation. Aust. J. Forensic Sci. 2016, 48, 454–468. [CrossRef] Diaz-Vilarino, L.; González-Jorge, H.; Martínez-Sánchez, J.; Bueno, M.; Arias, P. Determining the limits of unmanned aerial photogrammetry for the evaluation of road runoff. Measurement 2016, 85, 132–141. [CrossRef] Stáňa, I.; Tokař, S.; Bucsuházy, K.; Bilík, M. Comparison of Utilization of Conventional and Advanced Methods for Traffic Accidents Scene Documentation in the Czech Republic. Procedia Eng. 2017, 187, 471–476. [CrossRef] Mark Bateson Aerial Perspective. Available online: http://test.draganfly.com/pdf/Aerial-Forensics-V5.pdf (accessed on 22 August 2018). Liu, X.; Zhao, L.; Peng, Z.-R.; Gao, T.; Geng, J. Use of Unmanned Aerial Vehicle and Imaging System for Accident Scene Reconstruction; Transportation Research Board: Washington, DC, USA, 2017; p. 14. Liu, Y.; Bai, B.; Zhang, C. UAV Image Mosaic for Road Traffic Accident Scene. In Proceedings of the 2017 32nd Youth Academic Annual Conference of Chinese Association of Automation (YAC), Hefei, China, 19–21 May 2017; Volume 5. Stevens, C.R., Jr.; Blackstock, T. Demonstration of Unmanned Aircraft Systems Use for Traffic Incident Management (UAS-TIM); Texas A&M Transportation Institute: College Station, TX, USA, 2017; p. 73. Mekker, M.M. Evaluation of Consumer Grade Unmanned Aircraft Systems for Photogrammetric Crash Scene Documentation; Transportation Research Board: Washington, DC, USA, 2018; p. 22. © 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).