☰ ZKTeco Body Temperature and Mask Detection Reader Tested By: Brian Rhodes and Derek Ward, Published on May 26, 2020 While dedicated fever cameras emerged first, now tablet/kiosk fever detectors are ramping up. China's ZKTeco has been aggressively promoting such devices, even pitching it to IPVM members at the April New Products show. We bought and tested their device, the SF1005-V+ to see how well it performs., examining: • ZKTeco's accuracy compared to an IR thermometer. • What impact does height and distance have on temperature measurements? • Accuracy of mask detection. • How do ambient temperatures impact measurement? • How quickly can temperature measurements be made? • Impact of glasses, hats, and long hair covering the face. Key Findings Positives ☰ The SpeedFace SF1005-V+ performed well in several aspects: • When following ZKTeco's recommended (but highly restrictive) installation height, distance, and environment conditions, the thermal reader delivered accurate reads within their specified +/-0.54°F range. • Liveness detection was strong, as we were unable to get the reader to see a paper face as an actual human/user. • When constrained by their restrictive requirements or not having hair cover the forehead, elevated body temperature or failure to wear masks resulted in immediate notifications that prevented users from passing further. Negatives However, various errors were found in practical use testing: • We found temperature measurement errors when subjects stood too far away (past 18"), too tall (over 6' tall), or too short (below 4' 8"). • Sunlight, reflections, and dynamic ambient temperatures also skewed results, leading to 'fevers' falsely being detected when users did not have one. • While ZKTeco claims a half-second read time, our testing proved this to be false, with real-time measurements taking between 1.5 to 3 seconds per user. • While ZKTeco claims that the reader needs to be calibrated only once, we recalibrated between morning, afternoon and evenings. Our testing showed that if room temperature deviates ~4°F or more, the unit temperature deviated ~0.5°F or more. • Mask detection was simple to spoof by covering the mouth with a hand. Additionally, users with full beards were sometimes detected as wearing a mask. • Facial recognition effectively required separate user accounts for masked and unmasked faces. In the video below, we summarize our findings and issues noted, and offer our recommendations: ☰ Objection From ZKTeco After Testing While ZKTeco USA CEO Larry Reed approved the 5" unit prior to testing and supported our testing of the 5" unit throughout, at the very end, after we shared various problems found, Reed declared the 8" unit 'outperformed' the 5" unit. However, Reed has not yet explained in what ways the 8" unit does so. Before we started testing, we explicitly asked Reed about "test[ing] SpeedFace + 5 inch vs. 8 inch", questioning "Will you confirm the thermal/mask detection/face rec/palm print results will be valid for either model?" to which Reed responded noting "the body temp version (model SF1005-V+). Please double check he has this model number. If yes, we’re good to go." Pricing ZKTeco SF1005-V+ sells for ~$3,000 USD online. This pricing is lower than typical camera-based screening systems (e.g. Dahua, Hikvision, or Sunell) which generally sell for $12,000-15,000 USD. It is in a similar range though to other tablet / kiosk systems. Bad Reads / Misses Outside Recommended Distance ☰ In our tests, subject distance from the reader had a significant impact on correct temperature measurements. ZKTeco technical contacts recommended 18" measurement distance, but their documentation claims 40cm (~15.75"), even closer. With the recommended close distance of 18" to the reader, some users may feel awkward, with the reader too close or infringing on their personal space, making them prone to standing further away than recommended, skewing results. As an extreme example, on the left, a subject with elevated forehead temperature is properly detected at 18". However, on the right, he is measured around "average" body temperature from ~32" away at 98.51°F, despite elevated temperature shown on the IR thermometer. ☰ In the comparison below, we have a subject standing ~18", ~24" and ~32" away from the reader. Given how close users may feel to the reader, they may naturally back away from the device for some added personal space. The distances from the reader were taken from the subject's nose to the reader. Awkward Use For Tall/Short Subjects ZKTeco recommends that the terminal be mounted at 1.5m, ~4.9'. However, based on this mounting height, taller users may need to bow/crouch, while shorter users may require a step for proper measurement, also shown in their documentation: ☰ In our testing, properly face alignment for taller and shorter users was awkward, as much of the face was obscured by the panel's UI elements. Measurements of taller subjects were effectively similar to those of "average" height, but measurements of shorter subjects were ~0.5°F lower. ☰ Not ADA Compliant ZKTeco's device will not be ADA compliant for public building applications like courthouses, schools, or public transportation hubs (i.e.: Group Occupancies) due to lacking audio positioning prompts for users who are visually impaired. Unless manned with an operator that can assist users into the correct position, another method to measure body temp will be needed. In ADA-applications, temperature, face recognition, and palm recognition are all features that are likely not compliant, although options for the onboard fingerprint reader may be different if it is hung at permitted heights and clearances per ADA and Disability Laws. Missed Detection Possible In our tests, missed detections created by heating subjects foreheads prior to reading were possible, though less common than in camerabased systems (e.g. Dahua, Hikvision, Sunell). For example, the subject below was measured nearly 2°F lower than an IR thermometer, reported under the 100.4° fever threshold. ZKTeco does not include the resolution of the thermal imager on their data sheets, and we are awaiting ☰ Additionally, as seen in Hikvision and Sunell, hair covering the forehead consistently blocked detection of elevated temperatures, shown below: ☰ False Positives Caused By Backlight False temperature readings were not noticed with the reader installed in a stable environment with no windows or reflective material present within the view of the reader. However, when situated in front of a window with abundant sunlight, false reads would occur. ☰ Hats/Glasses/Masks Minimal Effect On Measurements ZKTeco does not have firm recommendations for removing hats and glasses when getting a temperature measurement, only saying that it will "have an effect." In our tests, subject temperature was effectively similar when wearing hats, glasses, or masks. ☰ Also note that outside of temperature, the system may be unable to properly recognize a user wearing a mask or other items, discussed below. In their marketing, ZKTeco claims 1,024 temperature measurement points, instead of focus on the forehead, canthus, or other areas, with the camera automatically picking the "best results." Calibration Impact On Accuracy ZKTeco's documentation vaguely states that the SpeedFace unit should be calibrated for ambient temperature "once" in a climatecontrolled environment. ☰ However, in our testing, we found that measurement results began to drift significantly (~0.5°F or more) if the ambient temperature changed by more than ~3-4°F, possible in lobbies as the day warms up or in locations without adequate climate control. For more on SF1005-V+ calibration and set up, the video below shows the options available for thermal detection. Mask Detection Mask detection generally worked well to detect users actually wearing masks. However, it could be spoofed intentionally or unintentionally in two ways in our testing: First, users were detected as wearing a mask simply by covering their mouths with their hands or arms. If the subject's temperature is below the alarm threshold and they can still be recognized, and then allowed entry, shown here: ☰ Additionally, thick beards were regularly detected as a "mask", even though the subject is clearly not wearing one. ☰ Finally, when enrolling users for facial recognition in ZKTeco's software, we had to enroll users wearing masks and without in order for them to be properly recognized. Users enrolled without masks were not recognized when wearing them, and vice versa: ☰ Strong Liveness Detection During testing, ZKTeco's liveness detection was strong and did not detect printed photos or photos on mobile devices as live humans: ☰ Longer Than Marketed Measurement Time ZKTeco claims that the SF1005-V+ is able to produce temperature readings around half a second. ☰ However, based on our testing, we found measurement times take between 1.5-2.3 seconds. We measured this time using several different people, performing more than a dozen attempts each. Reviewing Temperature Measurements Users will need to download and install ZKTeco's ZKBioSecurityV5000 to review and download images, as well as verify what individuals ☰ were triggered as over temperature, or "exempt" as the system describes. Physical Overview The ZKTeco SF1005-V+ is compact but includes the thermal and visual camera, white light, 5" touchscreen, IR Illuminators and fingerprint scanner. The device also allows for RS232, RS485, Wiegand, ethernet, and alarm I/O connections. We look at the physical construction of the Seek Scan camera and blackbody in this video: ☰ Version Tested Testing was conducted using the ZKTeco SF1005-V+ with firmware version ZAM170-NF-Ver1.1.26. This firmware was provided to us by ZKTeco during the testing. COMMENTS (10) Only IPVM Members may comment. Login or Join. Juan Arango 18 hours ago I dont understand if they claim to have a thermal camera built in the solution, why the final user needs to be so close to the device? also it will be good to know the integration with pre installed systems of access control. Brian Rhodes IPVMU Certified | | 17 hours ago also it will be good to know the integration with pre installed systems of access control.