How does the camera monitor 1.3 billion people?

From 2009 “Guidelines for the Construction of the Shared Platform for Sharing the Platform of Social Security Video Surveillance System in Hubei Province”, “Building Plan for the Monitoring System of Ping'an City in the North District of Qingdao City” in 2010, and “Three Years of Construction of the Video Surveillance System for Social Security in Huizhou City in 2012” In the several documents of Planning, we can see how this system works.

In China, the video surveillance system is also called "Skynet Project", which is usually divided into video acquisition-network transmission-data processing-business application and four-layer structure. In these documents, the camera's unit of measure is "road", and the procurement budget for each camera is between 30,000 and 50,000.

The first step in Skynet's project is to deploy a camera. These nodes are divided into a type of monitoring point and a second type of monitoring point.

One type of point mainly covers roads, squares, subway buses and other crowded areas. The second type of points are mostly important public spaces such as hospitals, schools and parks.

Taking Guangzhou as an example, in a Guangzhou video system construction plan released in 2013, the number of first-class points in the city reached 58,000, and the number of second-class points reached 173,000.

Different types of monitoring points have different requirements for the camera.

The newly set one type of point is mostly HD camera, which requires image resolution higher than 1280×720, and is monitored by the public security for 24 hours. The second type of point is relatively loose, the budget is about half of a class, and the real-time monitoring is carried out by the relevant units, not the public security.

Common surveillance cameras are divided into domes and guns according to their shape. The guns are usually installed indoors, corridors, entrances and exits, etc. where fixed monitoring is required. The advantage of the ball machine is that it can rotate continuously 360° in the horizontal direction, and more often appears in the open space, and the cruise can be rotated in time according to the specific situation.

These monitoring devices generate a large amount of image data every day, and how to efficiently transmit and store video data is also a technical difficulty of Skynet Engineering.

The key to video transmission is to reduce the delay as much as possible. The optical fiber transmission connecting the video optical transceiver is the mainstream solution today, and the delay from the front end to the monitoring center can be controlled within 2 seconds.

The storage and processing of the video is relatively simple, and the ordinary stone disk can be realized by accessing the stone disk recorder.

A camera with a resolution of 720 x 576 and a 70% change in motion rate produces approximately 0.25M of data per second and 211G of data per day. To achieve a video storage time of no less than 15 days, a camera needs to be configured with more than 316G.

Although the amount of data is not small, the cost of storing data is not high for each device with a budget of 4, 50,000. The 3,000 1T stone plate is enough to meet 10,000 cameras.

Skynet Engineering involves a wide range of links. In the list of video surveillance system equipment in Shibei District, Qingdao, there are as many as 54 types of equipment to be purchased.

From hardware, installation, transmission, to software, storage, and maintenance, every link is costly, and the video surveillance industry is also ushered in a period of rapid development.

In 2005-2015, China's video surveillance market has turned 14 times. Among them, the leading company is Hikvision, and its background of state-owned assets makes it safe in the security industry. In 2016, Hikvision achieved nearly 32 billion revenues.

Today, face recognition and intelligent analysis are starting to become standard for video surveillance.

With the breakthrough of the deep learning algorithm in the field of speech and image recognition in 2010, the recognition of faces by cameras began to become a reality.

Face detection is the first step in face recognition. Whether it is side face, occlusion, low light, or blur, today's technology can achieve millisecond detection.

Subsequently, the camera will select a high-quality one from the captured images, complete 106 key points such as nose and mouth contours, and finally compare with the ID card and other photo albums.