Is five seconds a lot or a little? To drink hot coffee is not enough, to attach a card and go to work - a lot. But sometimes even because of such a delay, queues are formed at the checkpoints, especially in the morning. Now let's fulfill the requirements for the prevention of COVID-19 and start measuring the temperature of all those entering? The passage time will increase 3-4 times, because of this, a crowd will appear, and instead of fighting the virus, we will get ideal conditions for its spread.
To prevent this from happening, you need to either arrange people in a queue, or automate this process. In the second option, it is necessary to read the temperature of a large number of people at once, without loading them with additional actions. This can be done by adding a thermal imager to the video surveillance system.and perform several actions at once: identify faces, measure temperature and determine the presence of a mask. We talked about how such systems work at our conference " Biometrics against a pandemic " and will tell you more about it under the cut.
Where are thermal imaging systems used?
A thermal imager is an optoelectronic device that "sees" in the infrared spectrum. Yes, this is the same thing from action movies about dashing special forces and films about the Predator, which beautifully paints an ordinary image in red and blue tones. In practice, there is nothing unusual in it and they are used quite widely: with thermal imagers they determine the position and shape of objects emitting heat and measure their temperature.
In industry, thermal imagers have long been used to monitor temperatures in production lines, industrial equipment or pipelines. Thermal imagers can often be seen along the perimeter of serious objects: thermal imaging systems "see" the heat that a person emits. With their help, security systems detect unauthorized entry into an object even in absolute darkness.
Due to COVID-19, thermal imagers are increasingly being integrated with biometric identification systems for access control. For example, thermal imaging devices integrated into BioSKUD (a comprehensive solution of Rostelecom, which is developed and produced in Russia) can measure the temperature of people, track movement and highlight individuals with high temperatures.
There are no mandatory standards for the use of thermal imaging systems in Russia, but there is a general recommendation from Rospotrebnadzor , according to which it is necessary to control the temperature of all visitors and employees. And thermal imaging systems do this almost instantly, without requiring additional actions from employees and visitors.
How non-contact temperature measurement systems work
The basis of the system is a thermal imaging complex consisting of thermal imaging and conventional cameras, which are packed in a common housing. If you are walking down the corridor and a plump two-eyed camera is looking in your face, this is the thermal imager. The Chinese pranksters sometimes make them in a white case and add small "ears" to make them look more like pandas.
Simple optics are needed for integration with "BioSKUD" and the operation of face recognition algorithms - for identification and verification of the availability of personal protective equipment (masks) for incoming people. Additionally, through a conventional camera, you can monitor the distance between people or between people and equipment. In the software, video information about the measurement results is displayed in a form familiar to the operator.
In order for the thermal imager to react only to the temperature of people, it already has an algorithm for detecting faces. The equipment reads the temperature from the thermal matrix at the right points - in this case, in the forehead. Without this "filter", the thermal imager would be triggered by hot cups of coffee, incandescent light bulbs, etc. Additional functions include monitoring the presence of protective equipment and maintaining distance.
Typically, at the entrance to the premises, thermal imaging systems are integrated with access control and management systems. The complex is connected to the server, which processes the incoming data using video analytics algorithms and transfers them to the operator's automated workstation (AWP).
If a thermal imaging camera detects an elevated temperature, then a conventional camera takes a photo of the visitor and sends it to the control system for identification with the base of employees or visitors.
Calibration of thermal imaging systems: from reference samples to machine learning
To set up and operate non-contact streaming temperature measurement, an absolutely black body (black body) is usually used , which absorbs electromagnetic radiation in all ranges at any temperature. It is installed in the field of view of the thermal imaging camera and is used to calibrate the thermal imager. In the blackbody, the reference temperature is maintained at 32–40 ° C (depending on the manufacturer), with which the equipment is “checked” each time by measuring the temperature of other objects.
It is inconvenient to use such a system. So, for the thermal imager to work correctly, the black body must warm up to the desired temperature for 10-15 minutes. At one site, the thermal imaging complex was turned off at night, and in the morning the black body did not have time to warm up properly. As a result, all those entering at the beginning of the shift had an elevated temperature. Later we figured it out, now the thermal imaging complex is not turned off at night.
We are currently developing an experimental technology that makes it possible to do without the black body. It turned out that our skin is close in its characteristics to an absolutely black body, and a person's face can be used as a reference. We know that most people have a body temperature of 36.6 ° C. For example, if you track people with the same temperature for 10 minutes and take this temperature as 36.6 ° C, you can calibrate the thermal imager by their faces. This technology, implemented with the help of artificial intelligence, shows good results - no worse than those of thermal imaging systems with blackbody.
Where blackbody is still in use, artificial intelligence helps in calibrating thermal imagers. The fact is that most thermal imaging systems involve manual installation of a thermal imager and its adjustment to the black body. But then, when conditions change, the calibration has to be done again, otherwise the thermal imagers begin to show the temperature with deviations or respond to visitors with normal temperature. Manual calibration is still a joy, so we developed an artificial intelligence-based module that is responsible for detecting blackbody and configures everything itself.
Is it possible to disguise yourself from algorithms
Artificial intelligence and machine learning are often used in contactless biometrics. The AI is responsible for detecting faces in the stream to measure temperature, ignoring foreign objects (a hot cup of coffee or tea, lighting elements, electronics). Well, the training of algorithms for recognizing faces in masks is a must have of any system since 2018, even before the coronavirus: in the Middle East, people cover a significant part of their faces for religious reasons, and in many Asian countries, masks have long been used to protect against flu or urban smog. It is more difficult to recognize a half-hidden face, but algorithms are also improving: today, neural networks detect masked faces with the same probability as a year ago without masks.
It would seem that masks and other personal protective equipment should have become a problem in identification. But in practice, neither the presence of a mask, nor a change in the hairstyle or shape of the glasses affect the recognition accuracy. Face detection algorithms use eye-ear-nose points that remain open.
The only "failure" situation from our practice is associated with a change in appearance with the help of plastic surgery. After plastic surgery, the employee was unable to pass through the turnstiles: biometric processors could not identify her. I had to update the photo so that the face geometry access would work again.
Possibilities of thermal imaging systems
The measurement accuracy and its speed depend on the resolution of the thermal imager matrix and its other characteristics. But behind any matrix there is software: the video analytics algorithm is responsible for identifying objects in the frame, for their identification and filtering.
For example, the algorithm of one of the complexes measures the temperature of 20 people at the same time. The capacity of the complex is up to 400 people per minute, which is enough for use at large industrial enterprises, airports and train stations. In this case, thermal imagers record the temperature at a distance of up to 9 meters with an accuracy of plus or minus 0.3 ° C.
There are simpler complexes. However, they can effectively cope with their tasks. One solution is to integrate the thermal imager into the metal detector frame. Such a set of equipment is suitable for checkpoints with a small flow of visitors - up to 40 people per minute. Such equipment detects people's faces and measures the temperature with an accuracy of 0.5 ° C at a distance of up to 1 meter.
Problems when working with thermal imagers
The non-contact measurement of the temperature of people in the stream cannot yet be called perfect. For example, if a person has been outside for a long time in cold weather, at the entrance the thermal imager will show the temperature 1–2 ° C lower than the real one. Because of this, the system can allow people with an elevated temperature to enter the object. This can be solved in different ways, for example:
- a) create a heat corridor so that before measuring the temperature, people adapt and move away from frost;
- b) on frosty days add 1–2 ° to the temperature of all incoming ones - however, this way those who arrived by car will fall under suspicion.
Another problem is the price tag of precision thermal imaging systems. This is due to the high production cost of the thermal imaging matrix, which requires accurate calibration, germanium optics, etc.