Hikvision AI Cloud cloud edge integration of intelligent video network applications

From the perspective of big data composition, unstructured data, semi-structured data and structured data are about 7:2:1. Among them, the unstructured data of the video is not only large in volume, but also the data form and processing method are very different from other data. How to solve the application dilemma caused by this differentiation?

We believe that the intelligent video network that is integrated by the cloud can exert its unique advantages.

Video perception cannot be without edge calculation

Edge computing and intelligent front-end, which are keen on the industry and research institutions, are an important part of building intelligent video networks.

Through the target classification, attribute extraction and lossless modeling of the video front end, intelligent analysis is more precise and faster, and agile response and intelligent control can be obtained in the field to achieve accurate perception and reduce the computational pressure of transmission and background, for example: intelligent recognition Access control gate system, intelligent face capture camera, etc. The built-in integrated smart chip will strongly support the implementation of edge computing, making the front-end camera more powerful, easier to build network structure, more reliable performance, and ensure that raw data is not lost. In addition, the edge calculation is not limited by the conditions of the computer room, and the cost is also much lower than that of the cloud computing.

Needless to say, to achieve full network intelligence, there can be no edge computing and smart front.

Video cognition cannot rely solely on the "cloud center"

The particularity of video data determines the way in which intelligent video networks can't take all the data on the “cloud center”. In a video network with large amounts of data and complex data, if all the video streams are collectively aggregated into the cloud center for storage and re-structured processing, it is unrealistic, uneconomical, and unscientific in terms of transmission, computing, storage, and application. It is easy to cause high delay, high load, high energy consumption and low efficiency. A flexible and healthy intelligent video network should be the cloud-edge-edge-edge node model of cloud-edge convergence, which is the Hikvision AICloud architecture (as shown in the following figure).

The edge node is the main sensing layer of the intelligent video network. It is the starting point for the classification of targets and events, the accurate perception of attributes and the extraction of features, and the execution point of business processing agile response. The edge node and the edge domain form an autonomous and resilient edge calculation. For the built-in non-intelligent HD camera, the intelligent calculation can be performed intelligently in the edge domain. The edge domain and the cloud center cooperate to realize application analysis such as big data fusion association, logical reasoning and value mining across time and space.

Intelligent video network, especially in areas such as safe city and urban traffic management involving a large number of video intelligent applications, should be based on the complexity of the business, the timeliness of response, the carrying capacity of the network, etc., around the user's own governance structure, according to "data On-demand aggregation, calculation of layered implementation, application of hierarchical response" ideas, rational and flexible system design.

Intelligent video network

Data aggregation on demand: due to need, who uses who

Whether it is the public security industry or other government industries, timeliness is the top priority. Low cost and high efficiency are one of the important principles of video network construction.

The edge domain is close to the edge node, which is the main functional area for receiving video intellisense. It is also the most vigorous, active and time-sensitive area in actual combat application, and requires high man-machine judgment efficiency. Taking the public security system as an example, from the perspective of provincial, municipal, district and county applications, the current application of video policing is at the district and county level, and suspected vehicle investigation, personnel trajectory tracking, face recognition, public security prevention and control, etc. All must be based on the principle of "fast". A lot of practice has proved that “who needs to save and who uses it” can better meet the needs of business applications.

Computational layered implementation: business needs, tiered computing

The layered calculation process is performed for different business needs, and the next layer calculation provides the basis for the upper layer calculation, and the upper layer calculation provides the accurate service for the next layer calculation. Through distributed computing and parallel collaborative processing to improve computing efficiency, effective use, rational allocation of computing resources, to achieve intelligent analysis results output more accurate, practical application faster.

The edge calculation of the front-end camera mainly performs structural calculations such as target recognition and attribute extraction, which provides conditions for the retrieval, comparison and analysis of the back-end; the edge domain provides services for the edge-side business response and the cloud center for big data analysis. Create conditions; the cloud center focuses on multi-dimensional big data fusion (such as video data, Internet of Things data, regular business data, etc.), providing intelligent cognitive applications such as logical reasoning and correlation analysis based on big data and the ability to calculate and manage elastic allocation. "Predictive, early warning, prevention" needs.

Apply rating response: hierarchically based on responsibilities

China's administrative divisions and government functions determine the depth and breadth of application of different departments at different levels. As mentioned above, the aggregation and calculation of data is designed to meet different levels of application content and rapid response requirements. In the case of safe city and urban traffic management, both intelligent video applications involve multi-level institutions such as provinces, cities, districts, and grassroots. In the face of complex business systems, hierarchical response is both an application requirement for real business and an actual follow-up. Working principle. For example, video policing applications can meet the business responsibilities of different police types in different departments according to different responsibilities, and implement video patrol prevention, receiving police, security prevention and control, crime investigation, etc.; for example, city management can be based on business requirements of different departments. To meet the needs of cross-departmental joint law enforcement, traffic management, community management, and public services.

As the application of big data analytics continues to deepen, the task assignment of applications at all levels is more intelligent. At present, some video-based services of provincial and municipal public security agencies are not limited to management, but rely on big data analysis to expand into cross-regional cases, target tracking and control, and visual command.

In order to realize the deployment and landing of the AICloud architecture, Hikvision has launched HacStack, HacDataX, HacMaster and other products to meet the needs of intelligent algorithm warehouse, computing storage resource pool, big data resource pool, and IoT intelligent application. A video-specific intelligent front-end product that provides an open API/SDK interface for third-party integration. The ever-increasing application cases in various industries have gradually confirmed that Hikvision AICloud architecture is a reliable system that effectively satisfies the development of video intelligence. The intelligent video network that is integrated by the cloud allows us to “see far and go further”.