AI technology is helping a large number of people who are returning to work, reducing the risk of exposure to the new crown virus.
Editor’s note: This article comes from WeChat public account The heart of the machine (ID: almosthuman2014) By the author, Synced.
The epidemic of new-type coronavirus pneumonia is continuing, and various provinces and cities in China are expected to face a large-scale rework on February 10. For the people and local governments, each party is about to face another test after the long holiday.
The outbreak has aroused great attention from various places. In order to reduce the risk of transmission, many streets and villages in Beijing have been conducting closed management for a period of time: entering and exiting communities need to register and check their body temperature.
For most people, they need to go out to buy food even if they are self-isolated at home. In recent days, people have always been exposed to a variety of body temperature detection methods when traveling. Thermometer … If there are many people, the measurement needs to be queued. Sometimes inspectors don’t know how to operate the new equipment, which invisibly increases the risk of crowd gathering.
But the good news is quickly coming: At some subway stations in Beijing next Monday, we will see new artificial intelligence-based solutions.
On Tuesday, an AI temperature measurement solution was launched on a trial basis in the Haidian Government Affairs Hall and some subway stations in Beijing. It can achieve “15 suspected high fever alarms per second” on the task of screening fever groups. The super high speed without affecting the traffic efficiency.
Dense crowd, sensorless detection
This new technology about “temperature” comes from contempt technology. With the unremitting efforts during the Spring Festival of the “Emergency Project Team to Fight New Coronaviruses”, which was set up urgently, an AI body temperature detection system named “Mingyu” was launched in time to help Beijing’s New Coronavirus epidemic prevention and control.
The new body temperature detection system can realize “non-sensory monitoring” during subway security inspections. It has no effect on passenger movements and does not need to remove hats and masks.
Different from the past, this AI temperature measurement system innovatively uses human body recognition + portrait recognition technology, combined with the infrared / visible light dual sensor solution. At the same time, based on the independently developed artificial intelligence platform Brain ++, the hat with masks has been completed. The optimization of the detection algorithm makes it possible to quickly identify in crowd-intensive scenarios, without the need to remove the cap or remove the mask from the identified person, while improving the traffic efficiency and avoiding the spread of disease.
Compared to the manual close-range detection method and the traditional fever screening system, Deafsight focuses its research and development on the accuracy of long-range temperature measurement and the availability of high-density traffic. The front-end infrared camera is used to identify high-temperature people in the flow of people, and based on the human body and face information of the suspected fever person, AI technology can help various public places to help staff quickly screen abnormal body temperature in high-density crowd scenes to achieve non- The contact-intensive human flow AI-assisted temperature-sensing detection has largely solved the need for traffic efficiency and risk control in open places after the epidemic has stabilized.
In terms of processing speed, the Ming Ming system does not affect the measured person’s actions at all, and can achieve high-speed detection of 15 people per second (camera acquisition frequency reaches 25 frames per second), and a system can cover 16 channels , Which basically guarantees the control of entrances and exits of a subway station. In this way, the channel with a large flow of people is assisted by the AI system, and only one staff member can control the scene, greatly reducing the risk of front-line staff members being infected.
In addition to its fast speed, it also has a larger scope of application. This system supports non-contact long-range temperature measurement of more than 3 meters, and people who suspect a fever will automatically alarm. Combined with the self-developed human ReID detection and retrieval technology, the system can help workers quickly screen the clues of fever personnel for further confirmation and medical observation.
Intelligent body temperature detection devices give people a little peace of mind when traveling on public transportation. According to Guangkui, equipment such as Beijing will continue to be promoted, and eventually it is expected to cover more scenes such as shopping malls and squares.
Quickly set up a working group and complete deployment in more than 10 days
It is surprising that the AI temperature detection system has actually gone through more than 10 days of research and development. Among them, the algorithm development and internal debugging of the screening system for suspected fever patients were completed before the third day of the new year. “Kuangshi Technology started thinking about it as early as January 20,” Liu Ruiwei, a senior product manager of the Kuangshi Technology Business Product Department, told the heart of the machine. “Subsequent new coronavirus outbreaksThe outbreak, the company quickly established a virtual working group. “
Soon, the number of this group increased to more than 100, and Inge, CEO of Detective, personally took charge, and its members were drawn from various deficient departments. Everyone worked overtime during the Spring Festival and quickly conducted a feasibility study. They successively solved the problems of outsourced hardware equipment debugging and algorithm construction. The entire system was officially launched on February 4.
A member of the defiance project team nervously debugging the temperature detection system.
In this process, the developers who stayed in Beijing spontaneously went to the company to work overtime, while colleagues in other regions contributed their own power in the form of remote development.
“Unlike previous product development, the AI body temperature detection system is very meaningful for the prevention and control of the new crown epidemic situation. Although it is during the Spring Festival holiday in the development process, everyone’s enthusiasm is very high,” Liu Ruiwei said. The time to complete the preliminary design of this set of equipment was in the thirties.
While despising the deployment of new equipment, the management committee of Haidian District and Zhongguancun Science City also started active communication with various technology companies, providing a lot of assistance and unified scheduling of resources.
By using the AI algorithm, you can recognize it by wearing a mask
Under the current epidemic conditions, the vast majority of travelers wear masks, which poses difficult requirements for face detection in deep learning systems. “The conventional security method of airport gates needs to take off the masks. In our system, the algorithm has been trained specifically for the situation of wearing masks and hats. It can accurately detect people wearing masks, thereby ensuring accurate temperature monitoring. Sex, there is no need to remove the hat and mask by the subject. “Liu Ruiwei said.
Traditional face image segmentation. The AI body temperature detection system needs to recognize a human face while wearing a mask to help the infrared sensor locate it. ThisNeed to make some improvements to the previous portrait segmentation technology-developers have reset the key points of face image segmentation.
In terms of load measurement equipment, infrared temperature measurement has the advantages of non-contact and rapid body temperature screening, which is suitable for the detection requirements of long-distance, large-area, and dense crowds. However, many infrared temperature measuring devices also have problems such as serious temperature drift and inaccurate testing. The despised developer said that in order to achieve accurate measurements, the thermal imaging body temperature measurement system uses a black body as the reference temperature source for temperature measurement. The black body is installed in the field of view of the camera. The latter measures the temperature of the black body and uses this as a reference to perform temperature measurement correction in real time to meet the requirement of high accuracy of human body temperature ± 0.3 ° C.
In addition, pure infrared thermal imaging equipment does not detect the appearance of people. It is not intuitive for observers, and it is difficult to determine the identity of suspected fever patients. At present, in many public places, most of what can be seen are infrared pure thermal imaging, and some do not even realize the temperature measurement display, which cannot achieve efficient management and control effects.
Another major advantage of the temperature-deficient body temperature detection device is the integration of infrared and visible light sensors. Infrared is responsible for temperature measurement and visible light is for identification. Dual-light fusion solves the problem of video stream picture alignment, clock alignment, frame rate alignment, etc. of infrared module and visible light module time. Through the integration of the two SDKs, the system can map the data to the corresponding face and body in the visible light image for the temperature monitored in the infrared image, realizing the immediate and intuitive result presentation, and warning the suspected fever person, Convenient for second hand examination for those suspected of fever.
In the Haidian Government Affairs Hall, people are trying out the “Min Ming” AI body temperature detection system.
According to the introduction, the despised AI temperature measurement system is also divided into a portable version and a platform version. “After the system was launched, not only in Beijing, but also in other regions across the country, there were many demands,” said Liu Ruiwei.
Although the currently deployed system achieves high-speed detection and individual identification, infrared temperature measurement still has problems such as sensitivity to the ambient temperature, and there are environmental errors in areas where the isothermal temperature difference varies widely. Developers said that in the future they will further cooperate with hardware manufacturers to improve detection results.
Deaf is also actively developing an independent sign-in system and a face access control system with a temperature detection function to achieve active and accurate protection in specific areas. In addition to existing camera hardware, infrared sensing devices and AI components can be added, Rapid deployment and delivery; future integration of software and hardware integration equipment, to achieve low-cost, fast deployment.
In all walks of life, many people are fighting the epidemic in their own way. It is believed that with the efforts of people, we will soon usher in the moment when the epidemic has dissipated.