Discussion on cutting-edge technologies and future changes in the security industry

On September 5, 2020, the third "China Artificial Intelligence Security Summit" will be strongly restarted in Hangzhou. (Originally planned to open on June 20, but postponed due to the epidemic)

The most obvious sign that a scientist has established a new school of thought is a world-class major research direction that can be equated with the scholar, such as the theory of relativity to Einstein and neural networks to Geoffrey Hinton.

What are these research directions in the security field? Who are the representatives?

No one knows the right answer.

The security technology innovation in the past decade has been chaotic, fragmented, and growing wildly.

The project-oriented technology deployment model has also meant that basic research has had little impact on the industry.

Driven by the tide of independent innovation and new infrastructure, cutting-edge technologies will come further to the fore.

So what are the trends in security technology in the next decade?

The answer given by the "China Artificial Intelligence Security Summit" is: digital retina of the city brain, three-dimensional machine vision, and federated learning of data modeling.

The world-class leading scientists in digital retina, three-dimensional vision, and federated learning are Gao Wen, Quan Long, and Yang Qiang.

Gao Wen, Academician of the Chinese Academy of Engineering, former President of the China Computer Federation

Quan Long, Chairman of CVPR, the world's highest-level artificial intelligence conference

Yang Qiang, the first Chinese chairman of the board of directors of the International Federation of Artificial Intelligence

From 2018 to 2020, the China Artificial Intelligence Security Summit invited Gao Wen, Quan Long, and Yang Qiang to attend the security forum for the first time to talk about the world-class cutting-edge technologies and the changes they have brought to the security industry.

The 1st China Artificial Intelligence Security Summit

Gao Wen, Academician of the Chinese Academy of Engineering and Director of Pengcheng Laboratory,
Urban Brain and Digital Retina

Academician Gao Wen

In March 2018, Shenzhen, Leifeng.com AI Nuggets held China’s first conference themed “Dynamic Face and Vehicle Recognition”. AI Security Summit.

This is the first industry event that brings together the chief technology executives of the five major security companies (Haida Yutian Network) and AI unicorns such as SenseTime.

At the summit, Gao Wen, academician of the Chinese Academy of Engineering and director of Pengcheng Laboratory, delivered an opening speech entitled "Urban Brain and Digital Retina".

He mentioned that smart cities have been discussed for many years, and "video surveillance + AI" has also become the research direction of many companies.

At this stage, making cities smarter through surveillance cameras is not just a single video retrieval and computer vision problem, but a series of huge system projects such as whether it can respond quickly when faced with massive amounts of information and emergencies, whether it can reduce the amount of calculation, and whether it can effectively identify and retrieve.

The shortcomings of the existing video surveillance system make it impossible to complete many complex tasks. Even after the large-scale penetration of artificial intelligence, the demand side often adds specific smart cameras and processing systems for some special purposes. Some special cameras are only used to recognize license plates, and some cameras are only used to recognize faces. This patch-style approach actually brings many problems.
In response to these problems, Academician Gao Wen proposed the concept of "digital retina".

The so-called digital retina is analogous to the human retina, which is different from traditional cameras and evenThe visual computing architecture is evolving and innovating so that it can more intelligently support the city brain and serve intelligent applications such as smart security and city precision management.

Specifically, traditional cameras simply compress the captured video data and upload it to the cloud for storage before performing analysis and recognition processing.

The digital retina requires high-quality video encoding and visual feature extraction encoding of the captured video on the camera side. The compressed and encoded video stream is stored locally and uploaded to the cloud on demand. All compact feature streams are synchronized to the cloud in real time, which can not only ensure efficient storage but also conveniently support big data query and analysis.

At the same time, it supports adaptive migration, compression, update and conversion of deep learning models for intelligent video encoding and feature analysis between end-edge-cloud.

In short, Digital Retina is a scalable end-edge-cloud collaborative visual computing architecture that includes video encoding stream, feature encoding stream, and model update stream.

It has only been three years since the concept was formally proposed, but it took nearly five years from the initial conception, early practice, to the construction of theoretical foundations. Even now, the technical framework of data retina is still being improved, but its impact will be subversive.

As Academician Gao Wen mentioned in an article in 2018:

my country has clearly stated that "by 2020, we will basically achieve full coverage, full network sharing, full-time availability, and full-process controllable public security video surveillance network applications." However, without major technological breakthroughs, tens of millions of cameras simply cannot achieve "full network sharing" real-time data aggregation, let alone "full-time availability" network analysis and identification. "Big data" cannot be transformed into "big data", and huge potential value cannot be explored. Digital retina is a feasible disruptive technology development direction to meet the above challenges.

Academician Gao Wen's understanding of the "digital retina" has continued to deepen as he has gained a deeper understanding of the problems existing in the video surveillance system in cities.

According to available data, Academician Gao Wen first began to think about the defects of cameras in cities and possible improvements in early 2013 (or earlier).

At the beginning of the new year of 2013, Academician Gao Wen was interviewed by People's Daily Online. He mentioned that modern cameras are densely distributed, but they still need back-end manpower to monitor them. Once a major case occurs, the video footage is often not very useful.

If the system can extract, mine and analyze useful information from the data when designing video encoding, it will not only save manpower investment in the later stage, but also play an emergency role.
It is reasonable to guess that he already had a vague concept at this time, but he was still thinking about how to do it specifically.

In October 2013, Academician Gao Wen published an article titled "Video Coding, Analysis and Evaluation in Smart Cities" in China Information Weekly. In this article, he systematically proposed for the first time his in-depth thinking on the problems of video surveillance in "smart cities". He pointed out:

1. The current monitoring system is designed for video storage and human-centric video tracking, rather than computer-centric automatic analysis. Therefore, it is impossible to rely on such a system to implement a smart city video system and perform automatic analysis.

2. Video technology in smart cities faces three major problems, namely high storage cost (too much data), difficulty in retrieval, and difficulty in re-identifying objects. These three problems boil down to two problems in essence: one is the encoding problem, and the other is the video analysis and recognition problem. Therefore, how to efficiently encode surveillance videos and how to analyze and retrieve them are two essential issues that must be considered.

3. There is a strange phenomenon in the academic world, that is, scholars who do video coding are not interested in video analysis; and vice versa, those who do video analysis are not interested in coding. The reason is that the former deals with pixels and image blocks, which belongs to the field of image processing, while the latter deals with image features, which belongs to the field of pattern recognition. It is like two cars running on two roads that are difficult to meet.

Fortunately, Academician Gao Wen happens to straddle these two circles. He has a significant influence in both the field of video coding and the field of computer vision (especially face recognition).

His students, Chen Xilin and Shan Shiguang, have inherited his legacy in the field of face recognition and have now become leaders in the field of international computer vision.

His students, Professors Huang Tiejun and Ma Siwei, have inherited his legacy in the field of coding and have also had an important influence in the international video coding field.

Academician Gao Wen's analysis of the essential problems existing in contemporary urban surveillance systems laid the foundation for his subsequent research ideas and directions for solving this problem.

About Academician Gao Wen
Gao Wen is a Boya Chair Professor at Peking University. He received his bachelor's degree from Harbin University of Science and Technology in 1982, his master's degree from Harbin Institute of Technology in 1985, his doctorate in computer applications from Harbin Institute of Technology in 1988 and his doctorate in electronic engineering from the University of Tokyo in 1991.

He worked at Harbin Institute of Technology from 1991 to 1996, at the Institute of Computing Technology, Chinese Academy of Sciences from 1996 to 2006, and at Peking University from February 2006 to date. He is an IEEE Fellow, an ACM Fellow, and an academician of the Chinese Academy of Engineering.

His research areas are multimedia and computer vision, including video coding, video analysis, multimedia retrieval, face recognition, multimodal interfaces, and virtual reality.