Hikvision Temperature Screening Tested
Hikvision has ramped up the promotion of its 'temperature screening' system, including their salespeople arguing for no blackbody needed. But how well does it work with and without the blackbody?
We bought and tested the Hikvision system and inside this report, we examine:
- Hikvision's accuracy compared to an IR thermometer
- How performance compares with and without the blackbody
- Hikvision's temperature measurement strategy and what it gains and loses from this approach
- How Hikvision misses 'elevated skin temperature' detections
- Which conditions triggered elevated skin temperature detections
- How software setup impacts performance and measurements
- How Hikvision's engineering recommendations compare with their sales and marketing
This is the 5th in our series, joining Dahua Thermal Temperature Monitoring System Tested, Seek Scan Thermal Temperature Screening System Tested, Sunell Panda Cam Body Temperature Measurement Camera Tested and Smartphone Thermal Camera Body Temperature Measurement Tested (FLIR / Seek). Upcoming reports on FLIR, ZKteco, and more are on the way.
On the positive side, our testing found that Hikvision generally produced temperature readings of subjects within ±0.5°F of an IR thermometer when properly calibrated and using the blackbody. Additionally, the readings were consistently statistical normal that provided an increased sense of comfort.
However, even with the blackbody, it consistently underreported elevated temperatures by ~0.8°F (e.g., 99.0°F reported vs 99.8°F with a US Hikvision default alert temperature of 99.5°F).
Moreover, without the blackbody, the camera's performance fell outside of its own specifications (±0.9°F), with missed detections and false alerts both increasing as no single offset properly compensated for all subjects measured, despite Hikvision salespeople's claims that they do not require a blackbody. We especially do not recommend trying to use the system without a blackbody as the increased variance will clearly create increased problems.
Additionally, like Dahua, Hikvision's temperature compensation algorithm aggressively raises low temperatures on faces detected to closer to body temperature, including subjects moving at high rates of speed or with their foreheads obscured by materials much cooler than human skin, such as hats or even cardboard.
Finally, Hikvision only measures the forehead, no attempt at the inner canthus, which we confirmed in testing and with Hikvision.
Increased Misses Without Blackbody
The Hikvision camera became less accurate without the blackbody, with both increased false alerts and missed detections.
For example, the camera missed elevated forehead temperatures both with and without the blackbody. However, the margin between detected temperature and temperature measured by IR thermometer significantly increased when using the camera alone.
Decreased Accuracy Without Blackbody
When using Hikvision's blackbody, we found temperatures to generally fall within their ±0.5°F specification after proper manual calibration. However, when attempting to calibrate the camera without the blackbody, measured temperatures varied widely, generally moderately beyond the non-blackbody specification if ±0.9°F (typically ±1.0 to ±1.5F). This variability resulted in false alerts when using Hikvision's averaging method, with some subjects measured higher than actual, triggering false alerts:
Only Forehead Temperature Measurement / No Inner Canthus Screening
Hikvision uses only subjects' foreheads for skin temperature measurement, found in our testing and confirmed by Hikvision engineering. The system does not use other areas of the face, such as the inner canthus, even if they are available and warmer than the forehead.
As a demonstration, the subject below warmed his left cheek to over 101°F using a hot water bottle, but the camera continued to measure him at ~97.7°F based on his forehead reading, highlighted by the red crosshair:
Second, in the example below, the camera measures cardboard which obscures the subject's eyes and forehead, measuring the subject at 95.2°despite being unable to clearly see his skin. Again, the red crosshair shows the camera attempts to measure the subject's "forehead" only.
This focus on the forehead can cause problems as it can be easily obscured. For example, simply having hair covering the forehead obscures a 102°F temperature:
Effects are similar when wearing hats, balaclavas or other face coverings.
In our discussions, Hikvision said that the camera previously measured other areas of the face, such as the inner canthus, but they found the forehead skin provided more consistent results than switching back and forth between the forehead and inner canthus (i.e., the inner canthus may be blocked by glasses).
Our concern, from testing, is that the forehead is going to have increased variability and error as people are entering from relatively warmer or colder areas (i.e., coming indoors during the summer or winter).
Elevated Temperature Methodology
To simulate fever, our subjects elevated the temperature of the forehead using a hot water bottle in the low 100s° F. The subject was then measured by the Hikvision camera, and finally by a handheld IR thermometer (Extech IR200).
While health and safety measures prevented measuring actual sick or coronavirus infected people, this approach is similar to what manufacturers use to prove their systems can alarm. Our approach differed by controlling the heated temperatures to realistic fever levels and verifying those.
Our methodology is conservative in that (1) we heated the entire forehead and (2) we had Hikvision measure the subject before we use the IR gun and in the intervening few seconds between Hikvision and the IR gun's measurement, the subject will, minimally, cool slightly.
Hikvision's camera uses an algorithm to estimate human body temperature based on skin temperature, similar to Dahua and Sunell. The cooler the object, the larger the offset between Hikvision reported temperature and measured surface temperature (using an IR temperature gun).
The chart below shows the difference between Hikvision measurements and surface temperature measured with a thermometer. Note that results were similar with and without the blackbody.
Hikvision's algorithm is similar to Sunell, but less aggressive than Dahua. While Dahua's offset shrinks to a fraction of 1° near body temperature, Hikvision's offset was never less than 2° in our tests, ranging from 86°F surface temperature to nearly 102°F.
However, unlike Sunell and similar to Dahua, Hikvision attempts to provide measurements of human temperature in poor conditions, discussed below. By contrast, Sunell simply displayed an under-temperature indication for objects below 95°.
Bad Reads Become More Normal Temperature
Hikvision, like Dahua, frequently raised displayed temperature to close to average body temperature, despite poor positioning or fast movements which make measurement difficult.
For example, even when the subject covered the top half of their face, obscuring the eyes and forehead (the warmest parts of his face), Hikvision displays a 95.2°F measurement.
With the subject riding past on a skateboard, he was detected and measured by Hikvision at 97.5°F. Shown below in iVMS-4200, the camera detects the subject, but he is blurred, moving so fast the bounding box appears behind him.
Similar effects were seen when running or biking, as well, similar to Dahua. By contrast, Sunell generally ignored these objects as under-temperature (turning bounding box blue).
Hikvision Recommends Automatic Adjustment / No Manual Calibration
Hikvision recommended to us that nearly all settings be left at default, with no need to use manual calibration (human temperature compensation) or manual ambient temperature adjustments unless readings are outside of the camera's specs (±0.5°F with blackbody, ±0.9°F without).
In our tests, based on this recommendation, manual calibration was still frequently required, with the camera most often reading subjects lower than actual temperature by as much as 1.2°F without blackbody, but closer to ~0.5-0.7°F with it.
We review these and other settings in this video:
Engineering Recommendations Vs. Marketing Claims
Hikvision's recommendations for temperature screening setup are similar to others like Dahua and Sunell, with subjects approaching in a single file, turning to walk past the blackbody, and then turning out of the scene.
Also similar to others, they recommend installing the camera indoors, in an area with "consistent temperature condition" and that subjects stay indoors for more than 5 minutes prior to measurement.
Contrary to these recommendations, a number of their marketing videos show cameras set up directly facing entrances, with strong backlight and opening/closing doors, and subjects measured immediately after entering.
However, Hikvision did not false alert on subjects entering from outside. As shown in the image below, the cursor showing measurement location is clearly located on the subject's forehead throughout, ignoring the hotter pavement behind him which triggered both Dahua and Sunell.
According to Hikvision, hats should be removed prior to measurement. While this is not documented in their brochures or setup guides, most of their marketing reflects this, with no subjects in hats found in their demo videos or brochures. However, the banner of the Hikvision global English site still shows two subjects in hats, shown here:
As shown above, hats block measurement, with the camera still attempting to measure the forehead, reading much lower than actual due to the hat covering. Hikvision says they plan to remove this image from their marketing. We will update accordingly if/when it is removed.
Live Monitoring Temperature Events
Hikvision's iVMS-4200 client displays temperature measurement events from the camera in real time, similar to other analytic events. There is no dedicated task for temperature screening.
The DS-2TD2636B-13/P is a similar size and form factor compared to other Hikvision bullet cameras, with the addition of the second thermal lens/imager.
The JQ-D70Z blackbody is similar to others in our tests, such as Dahua and Sunell, about 4" square by 7" long, and adjustable from 30°C to 50°C (86° - 122°F).
We review the construction of the camera and blackbody in this video:
Hikvision's thermal screening cameras are priced at ~$7,000-12,000 USD online, depending on resolution and lens. The blackbody calibrator is sold separately, another ~$3,000-4,000.
This pricing is similar to competitors which use blackbodies, such as Dahua and Sunell, both in the $10,000-15,000 range. Hikvision's camera pricing (without the blackbody) is similar to FLIR's A400/A700 models, which range from $8,000-12,000.
Upcoming China Fever Camera Shootout
We plan an upcoming 'China' 'Fever Camera' Shootout to more fully contrast the pros and cons of Dahua, Hikvision and Sunell.