There are three major areas where AI plays an essential role in thermal camera-based temperature screeners. The algorithm augments the hardware features to make the readings more accurate and relevant to the current requirements.
Thermal cameras have become an
integral part of businesses and organizations after COVID-19 began. The idea is that these cameras can identify people with elevated body temperature. But by now, most customers are aware that thermal cameras by themselves are not enough for fever detection. Temperature screening cameras now have an AI component that helps them serve their unique purpose.
According to Ara Ghazaryan, CTO and Chief Data Scientist at
Scylla.ai, there are three major areas where AI plays an essential role in thermal camera-based temperature screeners. The algorithm augments the hardware features to make the readings
more accurate and relevant to the current requirements.
"In brief, the solution is still hardware," Ghazaryan explained, speaking from the perspective of his company's solution. "But we have introduced a number of algorithm-based tweaks that enhance the abilities of the hardware."
Smart targeting
For a temperature detection camera to work effectively, the subject has to be within a particular distance. Also, the camera should be aimed at certain parts of the body, like the inner eye, for instance, to get the most accurate reading.
Ghazaryan explained that they make use of AI make sure the camera is able to focus on what the customer wants it to. Since its software-based feature, it's completely flexible and provides the customer with adequate options to customize.
"The algorithm chooses where to target within an area," Ghazaryan said. "Let's say we need to focus on the facial area. The algorithm helps the camera scan the different spots on the face and decide which spot to measure temperature from."
Multiple measurements
Fever detection can be more accurate if the camera can take multiple readings of the same person. This means that even if the first reading is not accurate because the person just came in from outside or a warmer room, an average of ten readings will provide better results.
"This helps you eliminate any room for error," Ghazaryan said. "Our solution does this. Moreover, after the system analyzes the different readings, it will give an alert if the temperature level is not normal."
Partially eliminates the need for black body
A black body is used for two primary reasons. First of all, it helps to correct each and every measurement by using a null reference. The second reason is that depending on the environment, the readings from a camera may vary. For instance, the same person may show different body temperatures at different times of the day.
"This is also taken care of by the algorithm," Ghazaryan said. "Although the blackbody is something that is much talked about, many customers find it costly and tedious to install and maintain. So if there is any solution that can help the customer avoid the use of a black body, it would really be welcome."
Ghazaryan added that his company relies on autocalibration. It relies on the analysis of the numbers that the system receives. The assumption here is that the majority of the people will have an average temperature, and the system tries to find people who exhibit body temperature levels that is not in the normal range. This makes sense because, eventually, the use of a temperature screening camera is not about finding someone's temperature level. It's about finding people who have a higher than `average temperature level. And AI helps to do just that.