In recent days, “the first flight of the Long March 7” has exploded the pages of major media. With the successful launch last night, this exciting Carnival has been stimulated to the highest point. While excited and proud, we should not ignore an important news: the birth of China’s first embedded neural network processor (NPU). This is a great news for the intelligent industry and has cross era significance for the field of video surveillance.

What is NPU? What does it subvert?

Before analyzing the NPU, you may have to bother to listen to the author’s wonderful man-machine war between alphago and Li Shishi. Many people are amazed and awed by alphago’s victory, but they don’t know much about its “weapon” of victory. In fact, the essence of alphago’s victory lies in its familiarity with the deep learning algorithm of human brain biological mechanism.

So what is deep learning? Deep learning is an artificial intelligence algorithm derived from the bionic research on the mechanism of biological human brain. Generally speaking, it means that the computer learns, judges and makes decisions by simulating the mechanism of human brain through deep neural network. At present, this method is used in face recognition and speech recognition. It should be emphasized that deep learning needs strong computing system computing power as support, and this computing power is inseparable from the development of processors, which is closely related to the NPU I want to talk about today.

At present, alphago uses CPU processors commonly used in other fields. Although this ordinary processor is universal, it is not professional. It is understood that in 2010, Google used 16000 processors just to train a deep learning neural network for recognizing cat faces. Not to mention defeating alphago in go. In the future, if artificial intelligence wants to use CPU to realize 100 billion neural networks like human brain, I’m afraid it’s not a long-term plan. Therefore, a processor that can miniaturize the deep learning system and can be used in the embedded system – NPU came into being.

What is the development of video surveillance based on NPU chip

Different from the CPU which is hard to run large algorithm model, the information processing capacity of NPU processor is 100 times or even 1000 times higher than that of CPU. It is reported that NPU adopts the “Data-Driven parallel computing” architecture, which subverts the traditional von Neumann computer architecture adopted by CPU. This data stream processor greatly improves the ratio of computing power to power consumption, and is especially good at processing massive multimedia data such as video and image, so that artificial intelligence can play a great role in embedded machine vision applications. According to Zhang Yundong, chief technology officer of Zhongguancun High tech enterprise zhongxingwei, if the data transmission mode of CPU is vividly compared to “thousands of troops crossing the single wooden bridge”, NPU turns on the green light for data in the unmanned 24 Lane in the early morning, which is not only efficient but also low power consumption.

In this sense, NPU chip is an indispensable part of intelligent hardware in the future. Only with it can we have a solid and reliable intelligent computing platform.

How does NPU lead the development of video surveillance?

Of course, the change brought by NPU can not only replace CPU, but also mean that the development of video surveillance industry has officially entered the intelligent era.

As the author mentioned above, NPU is good at processing massive multimedia data such as video and image, so the video monitoring industry will be the main field in which it will give full play to its strengths. As we all know, video surveillance is a relatively integrated system, covering a wide range. Coupled with the continuous improvement of the encoding and decoding ability of HD video, video surveillance produces exponential growth and complex data. To “decode” these complex data, we need more intelligent analysis methods.

It should be noted that installing NPU chip in the camera is like a brain equipped with IQ burst meter for “eyes”. It can continuously improve its IQ through learning and training, and gradually increase the types of recognition. After in-depth learning, it can record the interested events into the video stream in the form of digital tags. If we want to find some information, we only need to retrieve it intelligently in the background, rather than relying on the way of watching video manually. That means that in the future, we can search the content in the video as “long hair, glasses and satchel” in the same way as searching text.

It should be noted that in addition to real-time search, NPU chip can also store the photographed information in the code stream. When necessary, it can retrieve according to the characteristics at any time, or use the way of “searching for pictures with pictures” to find all similar pictures with one picture.

From the digital era to the intelligent era, video surveillance contains unlimited business opportunities. With the joining of NPU chip, it is like having a “customs clearance card” with skills. It is understood that NPU has been successfully industrialized in the field of video surveillance. In the next step, it will be widely used in embedded machine vision fields such as intelligent driving assistance, UAV and robot.

Conclusion: in the past, the industry looked forward to the development of artificial intelligence in the field of video surveillance, but it is regrettable that there is a lack of realizable hard “weapons”. Now with NPU’s love to join, it can be said that it is like God’s help in hardware. In the future, how to design products with more market prospects with the support of NPU depends on security enterprises.

Responsible editor: CT

Leave a Reply

Your email address will not be published. Required fields are marked *