Yuntian Lifei: Artificial intelligence technology can solve the problem of child abuse in kindergarten

Standing on the perspective of monitoring security, we will find an inexplicable phenomenon: video surveillance is installed in the kindergartens involved, and the teacher's child abuse pictures are clearly recorded. It is these pictures that are taken after the event, so that parents can understand the situation on the spot and also lock in the responsibility of the teacher. The problem is, these teachers know why video surveillance is still unscrupulous? Video surveillance, the most binding technology that should be used here, is like a chicken. So, how can we play the role of video surveillance? How to use technology to achieve timely discovery and rapid intervention?

In this regard, Chief Program Officer solve Shenzhen Li Fei Yun Tian Technology Co., Ltd. Wang Jun gave his view: "the use of artificial intelligence technology, combined with the existing video surveillance network wide coverage, we feel completely solve the problem of the kindergarten child abuse is possible In addition, Yuntian Lifei has begun to invest resources to develop corresponding solutions. It will be piloted in several typical kindergartens in Shenzhen in the near future. The key points for solving the two core problems of the problem are: the lack of blind spot monitoring in kindergartens, and the second Dynamic face recognition."

For the first point, Director Wang Jun believes that it is easier to do it. "As long as the government and the society work together, it is ok to increase the deployment of the camera in the kindergarten, but we must also pay attention to two issues: one is the resolution problem. It is best to use a camera with more than 2 million pixels. The image quality of many existing cameras in kindergarten is not high, let alone the accurate recognition of the face. According to our technical requirements, the face image of the child entering the lens, two The distance between the eyes is preferably higher than 40 pixels, so that we can accurately identify them." As the chief solution director of the current dynamic portrait recognition leading manufacturer, Wang Jun said that this is of course his enthusiasm. "The second problem is the on-site construction problem. Many manufacturers will ignore this problem. The traditional video surveillance is actually more. It is to record what happened at the time, not to collect faces, which leads to the background not being able to accurately define the identity of the person in the image, and it is impossible to accurately push the child to the parents. So, for on-site construction, Be sure to strictly control the position and angle of the camera, so that the camera is suitable for capturing high-quality face data."

What is more difficult to do is the second point, that is, the “emotional” recognition of dynamic faces, Wang Jun said. “At present, there are 6 billion face data in Shenzhen, Yuntian Lifei, adding 12 million face data every day. These data are constantly training the face recognition model, and the accuracy has increased from 92% to 96% in the past, so the identification of the face of a large number of children has been relatively mature. The difficulty is how to determine the identity of the child while still It can recognize the emotions on the child's face. Generally speaking, when children are abused, their faces are often accompanied by expressions of crying, fear, anger, etc. How to accurately identify the emotions behind these expressions is a great challenge for us. But the challenge is big and the power is great.” Wang Jun added with a smile: “Dr. Wang Xiaoyu, the former Snap chief scientist who recently joined our company, is further planning the company's algorithms and big data, and has helped us find a solution and training model. ”

“There are three steps. In the first step, you have to define emotions accurately. There are many kinds of emotions. Psychologists and philosophers have been talking about emotions for more than 100 years. According to the latest research, there are 27 people in humans. The most basic emotions, it is obvious that such classifications are too many and not practical. But just as the three primary colors of red, yellow and blue can evolve into ever-changing colors, we have been studying to find the most basic human core emotions. Last year I watched an animated cartoon about the children's minds: "The Mind Agents", which talked about the five emotions in the child's mind, called Lele, worry, fear, anger, and disgust; I think that The classification of children is very reasonable. All of the above, we finally locked in five basic children's emotions, namely joy, anger, sadness, fear, and hate. We use this as a standard to classify and train our algorithms."

Wang Xiaoyu, chief scientist of Yuntian Lifei, also said: "Our face detection algorithm is very robust to children's scenes. All faces detected will pass our age classification module, so only the face of the child will reach our system. These toddler faces will be sent to our emotion recognition engine. When the child is nervous or crying, it will trigger our face push alarm, which effectively reminds the parents to intervene."

Wang Jun went on to say: "The next thing to do is to collect data and linkage alarms. Artificial intelligence is trained with big data. It is very important to collect the child's face data. I think there are two kinds of face data for collecting children. Pathway, one is the mass collection of the kindergarten, and the other is the parent's upload. We will provide an app phone software for parents to use. This software not only provides a breakdown of the children's daily emotional classification photos, but also multi-dimensional statistics. The entry upload function allows parents to upload their daily facial expression data to the cloud, and our platform software in the cloud will automatically recognize and process. This requires a process, which may be biased at first, but with the increase of data, face recognition And the accuracy of emotional classification will increase rapidly."

"As for the linkage alarm, it is the last mile of the Long March, and the most important one kilometer. There is a contradiction here, that the platform software can not be frequently alarmed, but it cannot be alarmed, so there is a problem of setting the alarm threshold. These parameters need to be slowly adjusted in practice. However, once the warning signal is sent to the parents, the parents can directly retrieve the relevant video of the kindergarten on the mobile phone according to the alarm signal, define the situation and intervene in time.

"In short, in the future, we will build an emotional recognition solution for children's dynamic face like "end-cloud-end". The front-end camera of the kindergarten will complete the dynamic face collection 24 hours a day, and complete the emotion classification and intelligence in the cloud platform software. Identify, once the problem is found, send an alarm message to the parent's mobile phone in time to complete the intervention."

Wang Jun also said that in the parent-child garden, kindergarten and other occasions, the installation of monitoring should become an essential option, at least to ensure the evidence collection after the accident. However, the kindergarten is not a private area. It is a semi-public area covering teachers and many children. There is indeed room for improvement in the legislation of installing video surveillance in the kindergarten without any dead ends. The most fundamental thing is still to call for solving the serious contradiction between the supply and demand of early childhood education, changing the weak position of parents to implement supervision due to imbalance between supply and demand, solving the shortage of teachers in kindergarten teachers, and the uneven level of private schools. Through full competition, Forced all kindergartens to strengthen security control. Finally, Director Wang Jun said: "It is a difficult road to protect children with the power of science and technology. But as long as there is a road, don't be afraid of the road. Yuntian Lifei is willing to protect the children of the whole world through artificial intelligence technology. convoy!"

Guarding children to care for the future, Yuntian Lifei is already on the road. Through independent research and development of artificial intelligence technology, Yuntian Lifei can quickly locate missing children from the video stream and improve the detection efficiency and detection rate of children's lost cases. The first generation of dynamic portrait recognition Yuntian “Deep Eye” system independently researched and developed in the Shenzhen Public Security System was launched within one year, assisting in the cracking of more than 2,300 cases of various types and recovering many missing and missing children. Among them, the case of a three-year-old boy who was abducted for 15 hours on the eve of the Spring Festival in 2017 has been widely circulated in public security and society. Yuntian Lifei is using the power of artificial intelligence technology to make any criminal behavior invisible. With the power of high-tech “eyes” system, Longgang and even Shenzhen have become the safest cities in China and the world.

About Yuntian Lifei

IntelliFusion is a technologically innovative enterprise in the field of artificial intelligence jointly created by two “Thousand Talents” national experts. Break through the bottleneck of large-scale industrialization of artificial intelligence through cross-border innovation in processor design and computer vision. "Vision intelligence, core implementation", Yuntian Lifei has been highly recognized by social capital, including real estate, Songhe, Shenzhen Investment and other social capital investment of several hundred million yuan.

Yuntian Lifei's "Visual Intelligence and Machine Learning Processor Innovation and Industrialization" team integrated core technologies such as machine learning algorithms, processor chips, big data analysis, and ultra-multidimensional vector retrieval. It took two years to help Longgang Public Security build the world's number one. A set of dynamic portrait recognition system--Yuntian "deep-eye", realized "wide-city video surveillance, second-level face search", and assisted the public security to crack more than 2,300 cases in various cases, and retrieved many missing children. Shenzhen Longgang will be built into the world's first “safety demonstration zone based on artificial intelligence”. Longgang's model has been quickly copied to various districts in Shenzhen, as well as more than ten provinces and cities such as Beijing, Shanghai, Hangzhou and Xinjiang, as well as Malaysia and other countries; “Deep Eye” has also successfully served the 2016 West Lake Core Security Circle and Wuzhen Internet Conference of Hangzhou G20 Summit. The national double-day week main venue.

Yuntian Lifei pioneered the visual terminal chip + artificial intelligence cloud service to empower the technology industrialization platform of various traditional industries, and build an artificial intelligence industry chain to the civil city, smart business, smart park, smart medical, tourism, pension and other civilian fields. Expansion; In the basic technology fields such as artificial intelligence algorithms and processor chips, we continuously introduce top international experts and talents, and gradually form an artificial intelligence complex with artificial intelligence core technology barriers such as algorithms, chips, and data, and at the same time covering multiple industries. Yuntian Lifei is using artificial intelligence technology to empower all industries and gradually realize the safety, health, convenience and pleasure of human life, “making intelligence everywhere”.