Introduction: On January 16, the first "An Vision" technical salon was held in the demonstration hall of the Guangdong Public Safety Technology Prevention Association. The salon was hosted by the Guangdong Provincial Public Safety Technology Prevention Association, the Guangzhou Safety and Protection Industry Association and the Shenzhen Longgang Intelligent Audiovisual Research Institute. The technology and product experts in the security industry gathered to discuss how artificial intelligence is in the security industry. The engineering landing.
On January 16, the first "An Vision" technical salon was held in the Demonstration Hall of the Guangdong Public Safety Technology Prevention Association. The salon was hosted by the Guangdong Provincial Public Safety Technology Prevention Association, the Guangzhou Safety and Protection Industry Association and the Shenzhen Longgang Intelligent Audiovisual Research Institute. The technology and product experts in the security industry gathered to discuss how artificial intelligence is in the security industry. The engineering landing.
"Guiyan" and "Intelligent Brain" to create a smart city
National Distinguished Expert, Head of Shenzhen Branch of National Engineering Laboratory of Digital Video Codec Technology, Peking University, Research on Artificial Intelligence of Pengcheng Laboratory Li Ge, director of the center, delivered a keynote speech on "Digital Retina: 'Eye Eye' and 'Intelligent Brain' in Video Surveillance.
Video surveillance is an important part of public security. Video produces a lot of data, but it is often "without eyes." Li Ge believes that there are three major challenges in the challenge of media big data, namely, storage difficulties (compression problems), retrieval difficulties (pattern recognition problems), and identification difficulties (object re-identification problems).
In terms of storage difficulties, the growth rate of video compression rate is much lower than the data growth rate of video surveillance; in terms of retrieval, the exponential growth of image and video data poses a huge challenge to pattern recognition; In terms of identifying a particular object from a large data set, it is often difficult to include a series of visually apparently similar objects collected from different camera networks.
So, how to solve these three problems? Building a smart cloud image video data processing center is one of the solutions, but this method can make good use of the existing large-scale camera, but it also has many shortcomings.
For example, the accuracy of traditional visual perception systems in object detection, pattern recognition and scene understanding is not high enough.
So how do you deal with these problems now? LeeLeather believes that "cloud brain + digital retina" is the solution.
"Cloud" is a layered decoupled AI platform, this need not be repeated. What is a "digital retina"? Li Ge introduced that the human retina has two functions of image reconstruction and feature extraction. Image reconstruction is the fine coding of visual content. Feature extraction is the recognition of video stream. Although the eye can completely image the image, but the image is completely imaged. The brain's analysis of the image is based on the extracted features. Therefore, the effective use of video surveillance data in the city can also draw on the process of human retinal imaging and brain processing.
Li Ge believes that the future video should be divided into two "streams", namely video coding + feature coding. The technical framework of the future digital retina should be: video feature compact expression technology + efficient video surveillance transcoding technology. Li Ge also distances the tradition from searching for video collections, and uses dual-stream technology to transmit only the feature stream of each camera, which is no different from traditional results.
Professor Li Ge, national special expert, head of Shenzhen Branch of National Engineering Laboratory of Digital Video Codec Technology, Peking University, director of Artificial Intelligence Research Center of Pengcheng Laboratory. Doctorate in the United States Auburn University (Auburn University) in Electrical Engineering, University of California, Davis (Univ. Of California, Davis) electrical engineering postdoctoral. He has worked in a number of multinational companies, engaged in wireless mobile communications, mobile phone baseband chip algorithm design and video codec SoC design and other cutting-edge projects. As one of the leaders in technology and management, he has participated in the creation of two semiconductor design companies, which were acquired by two US Nasdaq-listed companies. As the project leader, he participated in many national, provincial and ministerial level scientific research projects, including: National Natural Science Foundation of China-Guangdong Provincial Government Major Special Project: High-efficiency expression, in-depth analysis and comprehensive utilization of video big data; Electronic Development Fund Project of Ministry of Information Industry: AVS Development and industrialization of standard core chips; major instrument projects of the Ministry of Science and Technology: development and application of ultra-high-definition video real-time analysis and enhancement instruments. So far, he has published many international papers in the fields of intelligent video processing and analysis, video data mining, etc., and applied for a number of domestic and foreign patents.
Three-dimensional prevention and control era, AI and big data combined to land new thinking
Radio and TV Safety Research Institute product manager Xie Yixing around "AI + big data empowerment public safety governance
Xie Yixing first introduced the development history and current situation of public security governance. He said that the 80-90s was the era of traditional governance, and digital governance began in 2000-2005. In 2006-2012, networked governance began. In 2013-2017, it was a synthetic governance. In 2018, it entered the era of three-dimensional prevention and control governance characterized by data. The combination of big data and artificial intelligence played an important role in this era.
Xie Yixing introduced that “AI+ Big Data” can play an important role in mastering behavior trajectories, establishing big data analysis models, monitoring major events, monitoring and early warning of key vehicles, monitoring in key areas, and empowering communities. The function of AI+ big data of Guangdian Express has been applied in the Xueliang project and the Longgang branch smart police cloud platform in Longgang District, Shenzhen, and achieved good results.
Xia Yixing, Radio and Television Intelligent Security Research Institute, intermediate level of communication technology Engineer, security industry 11 years of experience, in the security project design, project management, product development and other aspects of rich accumulation. The main person in charge, participated in and led the Shenzhen Longgang Branch Smart Police Cloud Platform, Guangzhou Tianhe District Zhigan Security Zone and other benchmark projects, and had a deep understanding of the overall development history of the security industry and AI empowerment public safety governance.
The accuracy of face recognition algorithm and the actual landing
Litu senior architect Li Na from the perspective of algorithm recognition accuracy to talk about the key path of landing. Li Na said, face recognition Divided into two factions, one group believes that face recognition faces bottlenecks, new application scenarios are limited, and another group believes that there is room for further improvement in the accuracy of face recognition in the future, and there are more scenes to be explored. "Fortunately, we The state's understanding of face recognition belongs to the latter. "
Li Na divides face recognition into three stages. Before 2015, it was the 1.0 era. At that time, the accuracy of the algorithm was at a disadvantage or basically flat compared with the human eye, and it was in the era of weak intelligence. [123 ]After 2015, face recognition entered the 2.0 era. At that time, the machine recognition could exceed the level of the naked eye by about 100 times. At that time, the 1:1 comparison, that is, the authentication and integration verification scene began to be applied. In 2016, face recognition began to have practical applications in security, 201After 6 years, there is a dynamic control application, which means that the system needs real-time comparison performance, which has high requirements for computing power and algorithms. The 3.0 era of face recognition is not only to identify, but to mine the relationship between face recognition data, so that the data will be considered and help the application to further land. This requires higher recognition accuracy. To this end, the clustering technology is introduced according to the figure to solve the problem that data cannot be reused. △ Li Na Li Na, senior architect of Etu Technology, Master of Mechanical Engineering, Harbin Institute of Technology, Senior Project Manager. He has worked in the mainstream manufacturers of security industry, engaged in video surveillance and application technology, power environment monitoring and Internet of Things technology and video intelligent application system design and landing plan planning. As one of the project architecture leaders, he has participated in and led several projects of more than 10 million levels, and played a key role in the company from analog monitoring, digital monitoring to network monitoring development and integration with new technologies. The development path has a deep understanding and experience.