University of Computer Studies, Mandalay Abnormal Driving Behavior Detection for Driver Assistance System Using YOLO Title Presentation Khaing Zar Myint Aung (MKPT-3773) Date: 9.5.2023 Outline • • • • • • Abstract Introduction State of Research Aim and Objectives Motivation Contribution • Overview of the Proposed System • Background Theory • Work Plan • Conclusion • References University of Computer Studies, Mandalay 2 Abstract • • • • • The ability to robustly detect abnormal driving behavior has the potential to limit traffic accidents and save many lives. Abnormal driving behavior that threatens road safety includes aggressive, anxious, nervous, and unstable driving. Any of these can lead to dangerous situations in traffic. Therefore, the system aims to provide a robust mechanism to detect such abnormal driving behavior. To handle the abnormal driving problem, driver assistance system have been developed. University of Computer Studies, Mandalay 3 Abstract (Cont’d) • • Abnormal driving behavior detection is based on the detection of phone, face movements, smoking conditions, drinking conditions while driving and drowsiness. The system proposes YOLO based abnormal behavior detection algorithm to help the driving system and to reduce traffic accidents. University of Computer Studies, Mandalay 4 Introduction • Traffic safety remains one of the main areas of study in vehicular technology. • Road safety continues to be a major concern in both industry and academia. • Nowadays, vehicles offer several safety and convenience features including partially or even fully autonomous driving. • Despite of that, the prevalent reason for accidents remains to be human reaction and abnormal driving behaviour. • The first step in mitigating such behaviour is to detect it in order to increase safety. University of Computer Studies, Mandalay 5 Introduction (Cont’d) • Abnormal driving is mostly caused by drunkenness, recklessness, and fatigue. • With the rising popularity of smart phones, mobile use is now the leading cause of death behind the wheel. • Normal driving is characterized by good speed control of the vehicle and the avoidance of sudden acceleration. University of Computer Studies, Mandalay 6 State of Research No Authors 1. Liu, C., Tao, Y., Liang, J., Li, K., & Chen, Y. In 2018 IEEE 4th information technology and mechatronics engineering conference (ITOEC) (pp. 799-803). Title Object detection based on YOLO network Method Summary Application Area Issues YOLO neural network used to analyze for research object. The experiment was based on the Darknet-53 network structure. Traffic signs dataset Training images: 1318 Testing images: 334 Total: 1652 images The trained model has better generalizing ability and higher robustness only with the degraded training sets. University of Computer Studies, Mandalay 7 State of Research (Cont’d) No Authors 2. Lin, J. In 2020 IEEE Internatio nal Conferen ce on Power, Intelligent Computin g and Systems (ICPICS) (pp. 420424). Title Integrated Intelligent Drowsiness Detection System Based on Deep Learning Method Summary Integrated Drowsiness Detection Systemcombined the heart rate monitoring system and the facial expression detection system using Resnet50 for classification Application Area Self-collected dataset for drowsiness detection and precaution system University of Computer Studies, Mandalay Issues The authors don’t state that the integrated drowsiness detection system is better in time efficiency than other methods. 8 State of Research (Cont’d) No Authors Title Method Summary 3. Zhao, L., & Wan, Y. (2019, December). In 2019 IEEE 5th Internation al Conference on Computer and Communic ations (ICCC) (pp . 21182122). IEEE. A New Deep Learning Architectu re for Person Detection Person Detection network model (PDnet) –improved the Yolo3 network model and the clustering function Application Area PASCALVOC dataset Training data: selected 5k photos Testing data: selected 3k photos University of Computer Studies, Mandalay Issues In complex scenarios, the algorithm can’t detect the object completely but still provides the possibility of low iou value. 9 Aim and Objectives Aim • The aim of the research is to develop detection algorithm that can be used in detecting abnormal driving behavior recognition system in real-time. Objectives • To assist the driving system and to reduce traffic accidents • To detect drowsiness and abnormal driving behavior • To become faster and more accurate in detecting objects • To evaluate the performance of the proposed system University of Computer Studies, Mandalay 10 Motivation • According to the latest WHO data published in 2020 Road Traffic Accidents Deaths in Myanmar reached 11,004 or 3.05% of total deaths. • The age adjusted Death Rate is 20.94 per 100,000 of population ranks Myanmar #71 in the world. • Deaths and injuries related to traffic accidents remain a major concern in Myanmar, which were mostly caused by human errors such as over speeding, negligent driving, drowsy driving and others. • One solution is to help driver to develop driver assistance system. University of Computer Studies, Mandalay 11 Contribution • The main contribution of the research is to construct the detection network architecture by using YOLO algorithm for abnormal driving behavior detection. University of Computer Studies, Mandalay 12 Overview of the Proposed System Training Process Testing Process Input Video or Image Input Video Labelling for Training Behavior Detection Image Preprocessing Results and Post Processing Proposed Network Model Trained Network Model Figure 1. Overview of the Proposed System University of Computer Studies, Mandalay 13 Procedures of the Proposed System • The procedures of the proposed system consist of the training process and testing process. • In the training process, firstly the user takes the video or image as input. • Then, the system makes a ground truth labelling by using the image labeller function in MATLAB for training. • In the stage of pre-processing, data augmentation is applied to the label image. • After pre-processing, the system extracts the features of image by using the proposed network model and train the image dataset. University of Computer Studies, Mandalay 14 Procedures of the Proposed System (Cont’d) • In the testing process, the input is obtained from the video or webcam. • The proposed system is captured these input frame and make behavior detection by using proposed detection model. • Then, the system will produce the results, make performance analysis and compare with traditional methods. • This research experiments on the datasets of primary sources or self-collected dataset. University of Computer Studies, Mandalay 15 Background Theory • Object detection is a phenomenon in computer vision that involves the detection of various objects in digital images or videos. • Some of the objects detected include people, cars, chairs, stones, buildings, and animals. • Object detection can be roughly divided into region proposals object detection algorithms such as RCNN, FAST-RCNN, FASTER-RCNN, and regression object detection algorithms such as SSD and YOLO. • Object detection network can be categorized into two main types: a two-stage network or a single-stage network. University of Computer Studies, Mandalay 16 Background Theory (Cont’d) Two-Stage Networks • The initial stage of two-stage networks, such as R-CNN and its variants, identifies region proposals, or subsets of the image that might contain an object. • The second stage classifies the objects within the region proposals. • Two-stage networks can achieve very accurate object detection results; however, they are typically slower than single-stage networks. University of Computer Studies, Mandalay 17 Background Theory (Cont’d) Figure 2. Overview of High-level architecture of R-CNN and Fast R-CNN object detection University of Computer Studies, Mandalay 18 Background Theory (Cont’d) Single-Stage Networks • In single-stage networks, such as YOLO v2, the CNN produces network predictions for regions across the entire image using anchor boxes, and the predictions are decoded to generate the final bounding boxes for the objects. • Single-stage networks can be much faster than two-stage networks, but they may not reach the same level of accuracy, especially for scenes containing small objects. Figure 3. Overview of YOLO v2 object detection University of Computer Studies, Mandalay 19 Background Theory (Cont’d) • From the point of view of computer vision, the abnormal driving behavior detection system is proposed by the traditional computing method and deep learning method. • Traditional computing methods take a long processing time for detection in real-time and the accuracy is lower than the deep learning method. • Although Convolutional Neural Network (CNN) based detection has high detection accuracy and faster computing time, it also takes a little bit long processing time per frame. • Although these approaches have solved the challenges of data limitation and modeling in object detection, they are not able to detect objects in a single algorithm run. University of Computer Studies, Mandalay 20 Background Theory (Cont’d) • YOLO (You Only Look Once) algorithm employs convolutional neural networks (CNN) to detect objects in realtime. • As the name suggests, the algorithm requires only a single forward propagation through a neural network to detect objects. • This means that prediction in the entire image is done in a single algorithm run. • The CNN is used to predict various class probabilities and bounding boxes simultaneously. • The YOLO algorithm consists of various variants. • Some of the common ones include tiny YOLO and YOLOv3. University of Computer Studies, Mandalay 21 Background Theory (Cont’d) • YOLO algorithm is important because of the following reasons: • Speed: This algorithm improves the speed of detection because it can predict objects in real-time. • High accuracy: YOLO is a predictive technique that provides accurate results with minimal background errors. • Learning capabilities: The algorithm has excellent learning capabilities that enable it to learn the representations of objects and apply them in object detection. University of Computer Studies, Mandalay 22 Background Theory (Cont’d) • An abnormal driving behavior detection system must alert the driver to prevent accidents in the driving state where the vehicle is moving at high speed. • So the detection algorithm must have high detection accuracy and real-time processing. • YOLO is a popular object detection algorithm that has revolutionized the field of computer vision. • It is fast and efficient, making it an excellent choice for realtime object detection tasks. • This algorithm is popular because of its speed and accuracy. University of Computer Studies, Mandalay 23 Work Plan Jan to Apr (2023) Conduct continuous, through literature review to identify gaps in existing algorithms and experts in the field Finding datasets for application area Prepare for seminar May to Oct (2023) Identify proposed algorithm based on research vision, plan, experimental results and literature review results. Prepare for seminar Nov to May (2024) Data preprocessing and testing with dataset and proposed algorithm Prepare for program Prepare for paper and then edit Prepare for seminar Jun to Dec (2024) Rewrite and rewrite a paper based on reviewer comments Prepare for seminar Implement the program Jan to June (2025) Implement the program Prepare for seminar University of Computer Studies, Mandalay 24 Conclusion • This research proposed abnormal driving behaviour detection by using the proposed detection model which is based on YOLO. • The system is developed a well object detection algorithm for detecting drowsiness and abnormal driving behaviour conditions. • In real world application, abnormal driving behaviour detection system can cooperate with alarm system and other. • The system can help prevent road accidents on a large scale and prevent loss of life as well as vehicle damage. • Car manufacturers can also think about adding the driver assistance system as a feature in the infotainment systems of their cars. University of Computer Studies, Mandalay 25 References [1] Liu, C., Tao, Y., Liang, J., Li, K., & Chen, Y. (2018, December). Object detection based on YOLO network. In 2018 IEEE 4th information technology and mechatronics engineering conference (ITOEC) (pp. 799-803). [2] Lin, J. (2020, July). Integrated Intelligent Drowsiness Detection System Based on Deep Learning. In 2020 IEEE International Conference on Power, Intelligent Computing and Systems (ICPICS) (pp. 420-424). IEEE. [3] Zhao, L., & Wan, Y. (2019, December). A New Deep Learning Architecture for Person Detection. In 2019 IEEE 5th International Conference on Computer and Communications (ICCC) (pp. 21182122). IEEE. [4] Hu, J., Zhang, X., & Maybank, S. (2020). Abnormal driving detection with normalized driving behavior data: a deep learning approach. IEEE transactions on vehicular technology, 69(7), 69436951. University of Computer Studies, Mandalay 26 Thank You