This article was excerpted from China News Weekly on March 5, 2018
The 13th Five-Year Plan for the Development of Modern Comprehensive Transport System, specially issued by The State Council, clearly points out that "to promote the intelligent reform of the transport industry, to promote the deep integration of cloud computing, big data, Internet of things, mobile Internet, intelligent control and other technologies with transport, to realize the digitalization and networking of infrastructure and transport tools, and intelligent operation and operation". With the 19th National Congress of the Communist Party of China clearly put forward the strategic goal of building a "powerful country in transportation", promoting the quality reform, efficiency reform and power reform of transportation development by intelligent technological means has become an important part of the development of urban new economy.
Give the traffic control department a "smart eye"
To build a powerful country in transportation and realize the historic transformation of transportation from large to strong, we must build transportation infrastructure with international competitiveness. Among them, the most important is the construction of intelligent road network monitoring infrastructure system.
By the end of 2017, the total length of expressways in China had reached 141,000 kilometers, according to the Ministry of Transport. According to the management standard of setting one camera every two kilometers, there are nearly 100,000 cameras only used for monitoring highway pavement (excluding toll stations).
However, China's high-speed road network monitoring still has the problems of inconsistent hardware equipment standards and insufficient intelligence. A large number of monitoring video images can only be observed and recognized by human eyes, and emergencies cannot be detected in time, and there are often missing reports caused by artificial fatigue. At the same time, traffic information cannot be interconnected, lack of intelligent sharing, a large number of real-time traffic data collected by the system is sealed or idle, it is difficult to form an effective management of the traffic network, as well as the prediction and early warning system for emergency events.
Tsinghua Tongfang Zheng takes cloud computing, big data and artificial intelligence application as the main line, and starts from the network layout of transportation infrastructure. It not only solves the operational management problems of the transportation industry, but also alleviates the pain points of ordinary People's Daily travel.
"At the present stage, with the mature landing of big data, cloud computing and artificial intelligence technologies, we have been able to form traffic problem analysis models through the ability of machine deep learning, and improve the management and service capabilities of smart transportation." Dr. Zhao Ying, general manager of Artificial Intelligence Video Technology Center of Tsinghua Tongfang Big Data Headquarters.
As a province with large population flow, Henan province is speeding up the construction of smart transportation and green transportation while improving its highway network. In order to solve high-speed congestion, reduce accident hazards and control exhaust emissions, Henan Provincial Department of Transportation took the lead in adopting Tsinghua Tongfang's artificial intelligence "Huiyan Da" system.
Tsinghua Tongfang "Hui Yan Da" artificial intelligence video analysis system.
Zhao Ying pointed out that the system, based on the theoretical system of machine deep learning and entropy change, has independently developed a series of basic algorithms and application algorithms. According to the video big data including the road traffic monitoring system, it can monitor the condition of video equipment and collect traffic operation parameters, and realize the intelligent prediction and early warning of road conditions.
"The Henan Provincial Department of Transportation is aware of the problems of resource waste and visual fatigue in the artificial reading and recognition of highway surveillance video, and is trying to find a solution to realize machine recognition. But the technology attached to upgrading surveillance hardware would require the replacement of existing road cameras, which would be a significant financial cost. Our 'Huiyanda' system is based on artificial intelligence technology, and can also be compatible with very old analog cameras, so as to meet the original intention of users not to replace cameras and make full use of existing highway resources to realize intelligent highway construction." "Recalled Zhao Ying.
In view of the difference between the video image speed and the real speed in the physical world, the system uses physical quantities that can be easily measured to build a macro entropy model, which greatly improves the detection accuracy of vehicle speed, and realizes the accurate detection of traffic events through the correlation between the group macro state and individual speed. To solve the problems of low accuracy and low value of image capture of traditional video equipment in traffic incident detection. At the same time, the system can predict the video blind area through the analysis of the road network running state, and identify the target through fewer feature points, so that the video can still achieve the target tracking in severe weather.
At present, Tsinghua Tongfang's artificial intelligence "Huiyan Da" system has realized real-time detection of 57 monitoring sub-centers and more than 3,500 video channels on 23 main expressways and more than 6,400 kilometers of expressways in Henan Province, solving video detection problems in complex scenes such as night, tunnels, bad weather and video blind spots.
"Traditional machine learning is based on shallow neural network for simple decision classification, while machine deep learning is built on the basis of deeper network structure and more application parameters. The network can be from a dozen layers to hundreds of layers or even deeper. Our system is built on machine deep learning." Zhao Ying said that Henan province has already applied the achievements of machine deep learning and has updated the version every year, which is now 3.0. As technology progresses and versions are updated, the false positive rate will be lower and lower throughout the network.
"We are also working to improve the algorithm, and by the time of the National Day Golden Week in 2017, the results were particularly good." "Said Zhao Ying. According to statistics, in the first half of 2017, the artificial intelligence "Huiyan Da" system, which has been put into use, accurately triggered nearly 400 congestion alarms, more than 1,000 parking cases and dozens of reverse traffic crashes, providing real-time road conditions information for nearly 10 million public during major holidays, saving millions of yuan of direct economic losses caused by congestion.
Following Henan, Tsinghua Tongfang's AI "Huiyan Da" system has been put into trial use in the road transportation network of 10 provinces, including Shandong, Inner Mongolia, Hunan and Qinghai. "China is the country with the most cameras." "Where there is a camera, AI can be effective," Zhao said. In addition to smart expressways, we have already made corresponding technical product reserves in smart areas such as urban roads, urban management, urban public security, underground corridor, and even public security, anti-terrorism and stability maintenance. This market will be explosive in the future."
Advancing smart transportation requires a bridge that connects cutting-edge technology to society's needs. "Tsinghua Tongfang is doing an 'uncompromising' thing. Together with Tsinghua University, we are building an industry-university-research platform to serve the urban development with cloud computing, big data, artificial intelligence and other high-tech technologies. This is the "site". We also take market application as a guide to nurture scientific research and innovation and help build an innovation-oriented country. This is the "sky high". Now we have started to cooperate with Tsinghua University and relevant government departments. In the future, we will use the big database collected by Huixingda to develop more intelligent services for the convenience and benefit of the people." Zhao Ying said.
