Company Touts AI-Powered Facial Recognition
“From a technical point of view, we still lag behind developed countries such as the United States [on artificial intelligence quality]. But in some areas, such as speech recognition, image recognition, facial scanning, etc., we may be at the forefront of the world.” – Miao Wei, the head of the Ministry of Industry and Information Technology (MIIT)
China is viewed as an increasingly formidable challenger to the U.S.’ current dominance in artificial intelligence, and has laid out blueprints to become the world leader in AI by 2030.
AI-powered facial recognition technologies in China have spurred the rise of a digital police state in Xinjiang, and are proliferating across China via the national Skynet and Sharp Eyes surveillance initiatives. Facial recognition is also used in areas such as ride-sharing, robotic package delivery, airport and college dorm security, and social credit schemes. One application that has drawn notable international attention was the facial recognition-powered naming and shaming of jaywalkers in Shenzhen. The following WeChat post cites directly from Yuntian Lifei, which operates under the Intellifusion brand and built the Shenzhen system now replicated across multiple Chinese cities. This excerpted post also describes their Skyeye system, which is capable of tracking individuals in real time, and its subsequent success stories in areas from human trafficking to theft prevention.
Through a set of systems that would utilize global video surveillance resources, Skyeye (天眼) can unconditionally provide real-time location tracking for any individual. This concept is common to the point of being unremarkable in science fiction films.
Yuntian Lifei was founded in 2014 by a team of doctorate degree holders who had returned to Shenzhen from the United States to start their business. We spent three years building a regional-level Skyeye system.
We combined face recognition built with deep learning with an image structure algorithm which combines many technologies such as processor chip technology, super multidimensional vector retrieval, and big data analysis to create a set of dynamic image recognition systems that operate from endpoints to the cloud. Today, this system now extends to Shenzhen’s airport, its subways, and ten administrative zones. Shenzhen’s 20,000-strong police force uses an app on the police cloud terminal; through this dynamic image recognition app, within two seconds, the police can track the activity of, locate, and view the criminal record of any suspect from the Shenzhen airport to the subway system, and across the region.
This system has been online in Shenzhen for nearly three years. It has assisted the public security department in cracking more than 6,000 cases involving economic investigations, technical investigations, network investigations, Interpol, human trafficking, counterterrorism, and smuggling.
[…] Many criminal cases cannot necessarily be shared publicly. There are several cases related to people’s livelihood.
The first case occurred the day before New Year’s Eve in 2017. A police station in Shenzhen received a report that a three-year-old boy had gone missing. Within two hours of receiving the report, the police successfully used Yuntian’s “deep-eye” dynamic image recognition system to quickly locate the criminal suspect’s trajectory and identity, and found that he had already brought the boy aboard a high-speed train bound for Wuchang. Through cooperation with the railway police, Wuhan police, and other local police departments, the boy was successfully rescued at Wuchang Railway Station in the early morning of New Year’s Eve.
From being abducted to being returned to the warm embrace of his parents, the whole process took 15 hours. In such an inter-provincial case, only such artificial intelligence technologies could have been capable of quickly sending a child back to his parents’ warm embrace before he even began to feel fear.
The second case took place in the first half of 2017 and also happened in Shenzhen. A police officer in a jurisdiction found that a young man was often stealing bicycles in the jurisdiction. Although he had a solid record of doing this, since the value of the case was too low, he could not be detained. This greatly affected the security situation in his jurisdiction.
The police sought our engineers for help. We tailor-made a system that was solely dedicated to the issue of stolen bicycles, and entered this young man as the first entry into this system. Two days later, we found that he was riding a new bicycle. This time the police did not deploy any forces, but instead conducted automated tracking within the system. The next day, he had taken the stolen bicycle to a repair shop, handed it to a repairman, and collected a certain amount of cash. This repairman was clearly a suspicious person, so he was the second person placed into this system.
We continued this way for four months. After we tallied up the cases in the area and the police had run their own investigations, a total of 318 criminals were arrested. This netted those working in this gray area of the industrial chain in one fell swoop.
The third case occurred at the end of 2016. The Shenzhen traffic police sought out our engineers. We built a pedestrian red light system at Lianhua Mountain Road at the entrance of Peking University Hospital in Shenzhen.
According to statistics collected by traffic police, in China, the proportion of fatal traffic accidents caused by pedestrians crossing on red lights accounts for 20-25 percent of all fatal traffic accidents. And at many large intersections, such as at the entrance of Peking University Hospital at Lianhua Mountain Road, the number of people who ran red lights every day tallied up to about a thousand people. This brought significant traffic safety hazards.
This set of systems was launched at the beginning of 2017. When the light turned red and the pedestrian set foot onto the crosswalk, when he reached the middle of the crossing, the facial recognition camera on the opposite side of the road would capture the jaywalking pedestrians and display not only his image but his identification information. Once you had crossed the road, to protect your personal privacy, some of the personally identifying information then becomes pixelated, and some digits from the national ID number become hidden. At the same time, the traffic police launched an exposing jaywalkers platform on their official website. Once they undertook privacy-protecting measures, they published the time and place of each time the person crossed on a red light, and associated it with the individual.
This is China’s first jaywalker system. After three months, the number of jaywalking pedestrians at this intersection dropped to about 8 percent of what it had been. Because of such results, the jaywalker system was replicated in Shandong, Hubei, and other provinces and cities in the second half of last year, and has become a phenomena in itself. And at the beginning of this year, media promoted rounds of discussion on the conflict between the artificial intelligence technologies and personal privacy, but in the end, public opinion was that the advantages outweigh the disadvantages.
These three cases are just a few small cases out of more than 6,000 cases related to social governance. Due to its disruptive effects across society, the system was quickly replicated in nearly 80 cities including Beijing, Shanghai, Qingdao, Hangzhou, Chengdu, etc. All had commercialized success in areas including Beijing’s Capital Airport, Beijing West Railway Station, and other administrative areas.
If you look closely, you will see that Yuntian Lifei cameras have begun to be installed in these areas. They have also been used in countries like Singapore, Malaysia, Vietnam, and other such Belt and Road countries. We have also participated in the core security work undertaken at conferences such as at the West Lake during the Hangzhou G20 in 2016, the Boao Forum, Wuzhen Internet Conference, Shanghai Summit, and other major international conferences. [Chinese]
Translation by Lisbeth.