The tide of reflow workers, standing on the first post of temperature measurement, became the key to sniper warfare.
Editor’s note: This article is from WeChat public account “ Heart of the Machine ” (ID: almosthuman2014), editor: Yu Yang
When the Xinguan epidemic became a large-scale public health event, security companies also began to join the epidemic prevention and control sniper war.
This is also a training ground for security and AI technologies. On the timeline of the race against time, companies quickly develop algorithms, design products, integrate supply chains, and finally deploy. After this battle, the intelligent security industry may accelerate differentiation.
The current epidemic is also a Spring Festival holiday. The security enterprise team is “up against the current”. With the cooperation of the government, the AI temperature measurement project is effectively scaled: quickly complete algorithm research and development, solve supply chain problems, and achieve floor verification … The story of the AI battle epidemic with emergency assistance was staged.
However, in this big test, the shortage of the supply chain and the delay of the start date have brought a lot of challenges to the actual landing of the AI thermometer. On the one hand, there is an urgent demand for public health facilities, and on the other hand, the epidemic There are many objective restrictions. What is the actual landing situation of the AI thermometer? What new opportunities will the security industry bring in 2020?
Emergency requirements under new scenarios
If you have set off on your way back, passing railway stations, airports, subway stations and other transportation hubs, it is not difficult to find that there is an additional camera around the security checkpoint.
Yes, this is a set of AI temperature detector, a tripod, a camera, and a notebook put together its basic form.
When a pedestrian passes by, his temperature will be shown on the display screen, and the sensor can pass without any sense when the temperature is normal; once a suspected high temperature pedestrian is detected, the system will automatically alarm and then the security personnel will perform a second temperature measurement.
In order to effectively respond to the high-traffic demand for epidemic prevention and temperature measurement, the AI temperature measurement project has a two-way blessing of infrared and artificial intelligence technology: infrared technology guarantees the need for temperature measurement at long distances without contact, and computer vision technology accurately recognizes Forehead area for quick positioning.
How to ensure efficient traffic and accurate measurementWen is a proposition for a new generation of “security + AI” companies to tackle the epidemic.
Baidu, launched the AI temperature measurement R & D work in the year of Jiu Jiu. Nearly a hundred people in various departments fought. On the evening of the night, the mask wearing monitoring model for the epidemic situation was completed, and the video data sample collection to detection model construction was completed within 3 days. The basic research and development work will be followed by a pilot deployment at Beijing Qinghe Railway Station on the fourth day of the new year to welcome the peak of passenger flow back to Beijing.
Despise, CEO Yin Qi took charge personally, and urgently set up a working group of more than 100 people. The preliminary design of the AI temperature measurement was completed in the thirteenth year. The algorithm development and internal debugging of the screening system were completed before the third day of the new year. After fighting for more than 10 days, the entire system was officially launched on February 4.
Together with the deep pupil of Geling, an emergency R & D call of the Beijing Municipal Science and Technology Commission on the second day of the new year, coincided with the idea of CEO Zhao Yong. A team of more than 30 people was quickly established in just 8 days Developed an AI temperature measurement system. The eighth day of the product has been “posted” in Beijing West Railway Station.
According to the heart of the machine, in order to catch up with the AI temperature measurement project, many employees live and live directly in the company, and some sleep only 2 ~ 3 hours a day … At present, no matter Baidu, Shangtang, Deaf, Gelingshen Hitomi, or Gaode Infrared, Hikvision, Dahua, Yushi and other companies, have launched AI infrared temperature measurement products.
From the market landing plan, the industry mostly uses dual infrared / visible light sensors, combined with infrared thermal imaging and face recognition, to achieve multi-target temperature screening in a non-contact manner, and the temperature measurement accuracy is mostly ± 0.3 ℃.
In terms of technical logic, it requires AI algorithms to do portrait recognition and tracking, that is, to accurately locate each face in the camera picture, and then calculate the face (mostly the forehead) with the infrared temperature dot matrix. temperature.
Details such as face recognition under masks, ReID (Pedestrian Re-Identification), and calibration between two cameras directly affect the effectiveness of each solution.
From the perspective of the industry’s positioning of AI temperature measurement products, the value it plays is still in the initial screening, and for people suspected of high temperature, manual secondary temperature measurement is still required to verify.
The reason why it is positioned as primary screening, on the one hand, primary screening + re-inspection can enhance safety from the process; on the other hand, it is limited by product accuracy and complex environment, which also has a certain impact on product availability.
In crowded scenes such as airports and train stations, once the crowd is in the cameraOutside the field of view, temperature detection is not possible. Therefore, when deploying AI temperature measurement products, certain personnel guidance is also required, or entrances and exits are set through the isolation belt to ensure that the temperature measurement is not missed.
Of course, there are still some restrictions on the initial screening program for fast launch.
The differences in technology and engineering between different companies bring different levels of temperature error. Although most solutions have an error of ± 0.3 ° C, some products still have errors of ± 0.4 ° C, ± 0.5 ° C. This is for availability Will have some impact. In addition, the working temperature of the infrared sensor is often 0 ~ 50 ° C. If the temperature is too high or too low, it will affect the temperature measurement accuracy. Therefore, in the northern areas of winter, outdoor may not be the ideal landing scene for AI temperature measurement equipment.
Although it cannot replace manual labor, the value of AI temperature measurement products is still important. In high-traffic scenarios, close-range security inspections are inherently very risky. Once crowd crowding is caused, the risk of security personnel’s work can be reduced through preliminary screening of body temperature.
Chen Tianbo, the product director of Geling Deep Pupil, also said that through the company’s test, the AI temperature measurement system has the same effect as the forehead gun. In his opinion, a good AI infrared temperature measurement solution must meet at least 3 conditions:
-
The first is to have a high temperature measurement accuracy.
-
Second, we must satisfy the high traffic flow in a non-inductive, non-contact, and non-cooperative manner.
-
Third, it should be easy to use and simple and intuitive for security personnel.
Once an element is not available, it may increase the risk of epidemic spread.
At the same time, AI temperature measurement solutions have also begun to spread to low-volume entrance and exit scenes such as buildings, gates, and access control. In addition to companies such as Shangtang, Deaf, Geling Deep pupil, etc., innovative Qizhi, Yufan Intelligent and other companies also launched AI access control systems with face temperature control.
As part of community control, such products are valuable, but there is an obvious problem. The face recognition rate under a mask is insufficient.
The deficient AI temperature measurement products currently do not support face recognition under a mask, which explains, “This time through AI image technology + remote temperature measurement, the core is to perform remotely more accurately in the case of crowds. Temperature test. ”
Shangtang Technology claims that when a person wears a mask, the face recognition accuracy rate can reach 85% when 50% of the nose is exposed. Although some access control manufacturers have added infrared temperature measurement modules, they do not support face recognition under masks. They still need to remove the mask to “brush face” or cooperate with access control cartoon. It can be seen that there are still some technical bottlenecks behind the initial screening.
The constraints of the supply chain
Under the epidemic, factories were shut down and logistics stopped, not onlyThere is a shortage of core infrared sensors, and even some accessories are difficult to find. Security companies have encountered non-technical problems. With the assistance of local governments, they have only taken this step.
Since the New Year’s Day (January 28th), security companies have successively promoted the pilot implementation of AI temperature measurement solutions. Baidu has launched a pilot project at Qinghe Station in Beijing and despised it in Beijing’s Haidian Government Hall and Mudanyuan Metro Station. In the pilot, Geling Shentong was put into use at Beijing West Railway Station.
After several days of trials, around February 3, each AI temperature measurement system has officially started operation. According to the data released by various companies, 15 to 20 people can be detected at high speed per second, and 200 to 300 people can pass the temperature measurement in one minute.
Contemplation claims that from February 4th to February 10th, its AI temperature measurement system went online for a total of 6 days, and the actual temperature measurement totaled nearly 5,000, of which more than 2,000 were measured on the first day of the resumption of work in Beijing on February 10. AI has begun to strongly support the return to work.
Along with the tide of rework, the urgent need for AI temperature measurement equipment from government units at all levels and the public sector. However, the other end is the severe supply chain shortage caused by the epidemic and the Spring Festival shutdown. The core infrared sensor has been included in the “wartime supplies” by the government, giving priority to the epidemic area, followed by the government, the public sector, and finally the enterprise.
Chen Tianbo, the product director of Geling Shentong, said that with the help of the government, AI temperature measurement equipment has been mass-produced. Currently it supplies about 50 sets per day, and after February 13, it can supply 80 to 100 sets per day. In addition to the Beijing area, many other provinces and cities have also raised demand, and companies will give priority to the public sector such as the government.
In order to shorten the supply time of products, security companies have hired “courier”.
Chen Tianbo said that the company in Beijing directly sent a car for transportation, and the operation and maintenance personnel also followed. When it was delivered, it was quickly installed and the air was sent outside.
“Now the landing speed is basically in hours. This morning (February 8) is still at Jinsong Subway Station (deployed), and at 11 o’clock at Xiaoying Subway Station (deployed), and then hurry to the next place “He said further.
At the same time, due to the shortage of the supply chain, it is difficult for companies to directly obtain core components such as infrared sensors, which has led to products such as AI temperature measurement, face brushing, and access control that are still in pilot or in short supply.
Innovative Qizhi claims that its current AI testingThe access control all-in-one machine has been tried out in some relevant government departments, and many government and enterprise customers are interested in buying. However, due to the tight supply chain caused by the epidemic, it will still take some time for product volume.
A sales manager of security manufacturer Yufan Intelligent said that with regard to temperature measurement components, the delivery cycle given by domestic companies is more than 6 weeks.
Will AI temperature measurement become routine?
After this battle, there is a voice in the industry that AI body temperature detection will become a basic service.
Baidu said that after the epidemic, the AI temperature measurement system can still be deployed in personnel-intensive places such as transportation stations, medical institutions, and various service centers. Pay more attention to relevant capacity building.
Chen Tianbo, product director of Green Eyes, said that while developing AI temperature measurement equipment, the company also extended the AI temperature measurement capabilities to existing industry scenarios, such as face-access control, building ladder control, and iterating the existing product.
However, due to the urgency of demand brought about by the epidemic control, the core supply chain is scarce, and AI temperature measurement equipment is mainly supplied to the government and other public sectors. Cost has become a neglected issue.
The market price of high-flow AI temperature measurement equipment used in airports, railway stations, subway stations, etc. often starts at 10,000 yuan, and even the equipment integrated into the face brush access control needs 10,000 yuan.
A sales manager of Yufan Intelligent said that the total price of software and hardware for facial recognition gates with AI temperature measurement is 12,000 yuan. The ordinary face-washing brake only costs two to three thousand yuan, and the price is several times higher.
Innovative Qizhi also said that currently various components, especially infrared temperature measuring devices are in short supply. On the whole, the ten thousand yuan level is already a reasonable price.
If AI temperature measurement becomes a basic service, such a high price will become a key issue for its scale.
This is not only related to the current shortage of the supply chain, but also to the program itself. For example, in order to ensure the accuracy of infrared thermal imaging, deaf eyes and Dahua have adopted the “black body” solution, and the real-time temperature correction of the black body is used to ensure the accuracy of the camera’s temperature measurement.
In contrast, Shang Tang and Geling Shentong have adopted non-black body schemes. Chen Tianbo said that it combines infrared thermal imager to measure temperature, ambient temperature, human body temperature and face temperature to form a unique algorithm, which can also control the accuracy to ± 0.3 ° C. Software guarantees accuracy as the market expands, Or will help form a more competitive price.
Besides the price, as more and more AI temperature measurement schemes appear or are integrated into existing products, the phenomenon of a rush of the industry has already appeared, but do you need such products in every scenario? What is the real need? The industry should look more rationally at the “new opportunities” that the epidemic has spawned.
There are also rumors that some manufacturers use industrial infrared sensors for human body temperature measurement, and chaos in the industry has already appeared.
Accelerate the differentiation of “AI security”
AI infrared thermometer, epidemic-proof outbound robot, AI diagnostic system for new coronary pneumonia, disinfection robot, epidemic-proof drone … AI is playing an important role in the epidemic.
In retrospect, the epidemic is more like a technical training ground. It tests the ability to develop algorithms for specific problems, quickly migrate solutions, integrate supply chains, and deploy capabilities. It tests the comprehensive capabilities of AI companies.
Although it is not a severely affected area, the epidemic has affected the business progress of AI companies to varying degrees. Under the epidemic situation, who can quickly develop a product that truly solves the problem and has withstood the test of the market, rather than blindly tapping hot spots, can occupy an advantageous position in the industry competition. From this point of view, the epidemic situation may accelerate the differentiation between AI companies to some extent.
Back to AI temperature measurement products, in addition to the participation of AI companies, there are also many security companies. Hikvision, Dahua, and Yushi have all unveiled intelligent temperature measurement systems, and they have a wider range of landings. Its supply chain integration capabilities and channel capabilities are yet to be strengthened by AI companies.
The increasingly mature artificial intelligence is becoming a special group of “retrograde” on the front of the fight against the new crown pneumonia epidemic.
Whether it is to accelerate the research of the epidemic in the early stage, or to assist the front-line epidemic diagnosis and treatment and on-site prevention and control, as well as the construction of the back-end information platform, artificial intelligence technologies and solutions have already appeared and achieved practical results.
