Artificial intelligence is being implemented in the security industry, but the lack of computing power is hindering its development

With the continuous improvement of technology and its application in various fields of social life, artificial intelligence is no longer regarded as a kind of "magic", but has become a productivity that empowers all walks of life in society. Baidu founder Robin Li said frankly that at this stage "artificial intelligence is no longer about being cool, but about how to promote and implement it in a down-to-earth manner."

If AI is the steam engine and the internal combustion engine, then AI computing power is "coal and oil" - AI applications are built on computing power. Without computing power to "burn", AI will be a castle in the air. A global AI computing power supply and demand research report by OpenAI shows that the demand for AI computing power doubles every three and a half months. From 2012 to 2018, the global computing power demand has increased by 300,000 times.

Lack of computing power hinders the development of smart cameras

Transportation, retail, logistics, smart cities... the traditional security industry is expanding into more dimensions because of artificial intelligence. However, the current AI computing power is stretched to its limits and cannot meet specific computing needs. Restricted by factors such as computing power, algorithms, and technical standards, the rapid processing and in-depth use of security surveillance videos still requires a lot of manual analysis.

The lack of computing power has also led to a slower-than-expected rate of popularization of smart security cameras. Currently, most common smart cameras are used in basic monitoring scenarios. To perform real-time identification and analysis of multiple targets in complex scenarios, the terminal computing power is not yet able to support it. Therefore, the current cameras can be regarded as front-end intelligence rather than intelligent front-end. Of course, the video can also be sent to the cloud for processing, but this will bring problems of high latency and data security.

"The future market will definitely be a competition of data scale and computing power," said Zhang Qi, director of OpenPOWER product marketing at Inspur Commercial Machine Co., Ltd. Whoever can solve the problem of insufficient computing power while better reducing power consumption and costs will be able to take the lead in security under the wave of AI.

Combining multiple technologies to alleviate problems

The problem of insufficient AI computing power is being solved by relevant companies. For example, an integrated deployment solution that combines cloud computing, edge computing, and device-side computing can alleviate the problem of insufficient camera computing power to a certain extent.

Whether AI is applied to consumer or industry scenarios, its future computing power will be multi-point collaborative. This requires different solutions to different problem scenarios, and these solutions need to be able to solve heterogeneous adaptation and migration capabilities for better forward compatibility.

For example, in some To B scenarios, the camera is additionally equipped with AI computing power to form a "fat terminal", and image recognition is performed locally through peripherals. This reduces latency while alleviating the need for network connection and capacity.

The design of these solutions is essentially to solve a problem - to allocate the tasks that need to be completed by the terminal and the cloud. However, the industry has not yet reached a consensus on whether AI computing power should be deployed more in the front end or the back end.

Putting the right computing power in the right place

To solve the computing power problem, the security field needs the coordinated development of cloud, edge and end, and put the right computing power in the right place. Of course, computing power does not exist in isolation, and other factors will also affect the use of computing power. For example, if storage optimization is not performed, the actual computing power that the chip can provide will actually be much lower than the theoretical value.

So, how to allocate computing power and balance other influencing factors? Hikvision's AI Cloud architecture may provide some ideas: using the cloud computing center to flexibly allocate computing server and storage server resources, and scheduling intelligent algorithms and big data algorithms on demand.

As the overall development of artificial intelligence technology is still in its early stages, the smart security industry is also facing great uncertainty, which also provides a huge stage for various security companies. What kind of wonderful performances will be presented on the stage? It is exciting!