2018 artificial intelligence security six development trends

In this context, companies must take a step to see the hundred steps, and the future trend of the industry under the new technology, and this will determine the company's product cycle, the final victory of industry competition. By this, Lei Feng Network summarizes and looks forward to the six development trends of the security industry in 2018, in order to attract readers. (for discussion only)

First, special chip innovation, AI landing security is more rapid

176 million units, this is China's current video surveillance camera.

Driven by major national security projects such as “Safe City”, the “Skynet” project was successfully rolled out from the city to the countryside. At the same time, people are at a loss as to the amount of mega-scale video data generated by massive cameras.

The "into" of AI has solved these problems. It can transform the video image content into structured data that clearly expresses the target attribute, and then intelligently analyze through the deep learning algorithm, effectively improving the data processing efficiency and deep data mining.

Although the AI ​​is more "hands-on" in handling massive amounts of data, its "worth" also makes it impossible for a security vendor that sells a camera to climb. Fu Liquan, chairman of Dahua Co., has publicly stated that price is one of the biggest obstacles to the AI ​​technology landing security industry.

This view has also been recognized by many security executives. They revealed that a common camera with face recognition and other functions increased the cost by 3,000 yuan. For example, NVIDIA's Jetson TX1 chip, which is used by Hikvision, is priced at more than 3,000 yuan, and the average price of a "squat" product is around 5,000 yuan, while an ordinary IPC is only about 1,000 yuan.

From this point of view, one of the important factors surrounding the smooth progress of AI landing security is: the high cost of AI chips.

The chip largely determines the overall functionality, technical specifications, stability, energy consumption, and cost of the security system. In the past few decades, the competition on the chip computing power has nothing to do with Chinese manufacturers. It has been “monopoly” by several major foreign manufacturers. Under the temptation of technology, security giants such as Hikvision, Dahua, and Yushi Technology have cooperated with Nvidia. Chip vendors such as Intel maintain close ties.

By 2018, this "monopoly" situation may improve.

Nowadays, the research and development of Chinese manufacturers around AI chips are in full swing. In addition to NVIDIA's general-purpose GPUs, there are currently smart chips developed for the security industry, such as the deep-tech DPU (FPGA), Cambrian AI server chip (ASIC), etc.; In addition, domestic companies that started with computer vision technology, such as Shang Tang Technology, Face++, Yun Cong, Yi Tu, etc., also extended their downstream chip links with their own technology accumulation.

From this point of view, the price of embedded AI chips suitable for smart cameras in 2018 is expected to decrease, thereby increasing the penetration rate of smart cameras. Of course, foreign chip manufacturers must have realized this, maybe next year's battlefield will move from this year's domestic algorithm war to next year's international power war.

Second, "cloud-side integration" will become the mainstream choice for security and intelligence

Back to the first point, because AI chips are expensive and subject to cost factors, in the past, security vendors used more back-end smart solutions to implement feature extraction and personnel retrieval through back-end intelligent NVRs.

But in the long run, cloud-edge integration is an inevitable trend. During the Shenzhen Anbo Conference this year, Hikvision President Hu Yangzhong publicly stated that the injection of AI power into the edge and the empowerment of edge intelligence is the general trend. By uniformly dispatching decentralized smart device resources such as IPC and NVR, the data source is provided nearby. Video-based sensing data, real-time pre-processing, storage and transmission services, while improving business agility, real-time and system reliability, share the pressure of concurrency brought by massive data to the central node.

The camera transmits 7*24 hours of uninterrupted real-time video stream. When the number of front-end cameras reaches a certain scale, the pressure on transmission and storage is enormous. In the front-end capture + back-end smart mode, the image stream is transmitted, and the bandwidth is occupied only when the face is captured, which greatly saves bandwidth and storage resources;

In addition, because the principle of cloud edge combination is to pre-set the intelligent algorithm, through the edge calculation, the pressure of the image capture of the application such as face recognition is distributed to the front end to liberate the computing resources of the center.

According to Luo Yi, vice president of business development of Etu Technology, the front-end intelligent design idea is that when the network communication bandwidth is limited, the manufacturer can only compress the video in order to save bandwidth, and in this process, the information must be subject to Damaged. For the customer, if the bandwidth is sufficient and the transmission cost is low enough, the most reasonable solution is to capture the required elements (structured video) in the original video and then transmit them back.

It is worth noting that today's cloud-side integration has not yet formed an optimal solution, which is a direction worth considering.

Third, the digital retina upgrades the urban brain

At this stage, by monitoring the camera to make the city smarter, not only a single video retrieval and computer vision problems, but whether it can respond quickly and reduce the amount of calculations in the face of massive information and emergencies. A series of huge systems engineering that can effectively identify and retrieve.

The shortcomings of the existing video surveillance system make many complex tasks impossible. Even after the large-scale infiltration of artificial intelligence, the demand side often adds specific smart cameras and processing systems for special purposes.

Some dedicated cameras are only used to identify license plate numbers. Some cameras are only used to identify faces. This patched method actually causes a lot of problems. We call it "one-to-one mode."

Then the digital retina came out.

The camera carrying the digital retina needs to do two things: first, coding; followed by the latter identification, extracting the required information. The digital retina and the human eye have both image reconstruction (finely encoded visual content) and feature extraction (for recognition understanding). How is the digital retina finally realized?

In principle, this section combines efficient, surveillance video coding (compact representation of video features) and features in a compact manner. With that, apply it to it and transfer it to the cloud to have digital retina functionality.

The entire digital retina actually contains three core technologies: scene video coding based on background model, compact expression of video features, joint optimization of video coding and feature coding.

In the future, when the one-to-one mode becomes a one-to-many mode, at the same time, after the technical standards in the one-to-many mode are in place, the technical standard can be further optimized and then embedded in the product to be practiced and applied as soon as possible. Cities will be easier to govern.

Fourth, the security C-end market potential began to release

"The explosion of the security C-end market is here!", this sentence has been mentioned in the past few years, many times. Since 2018, this market may have really started to heat up, and several major reasons will trigger the gradual warming of this market:

The popularity of smartphones and the upgrade of communication networks have provided a developmental soil for civil security systems based on mobile Internet;

In addition to Haikang, Dahua, Dongfang Netpower and other security companies, 360, Xiaomi, Baidu and other giants also participated in the launch of home camera products;

After 90, it has gradually become the backbone of society. This group is the first group in China that is easy to accept the new things on the Internet;

The price bottleneck is broken and AI is further mature

Among them, the fourth point is the most important. The previous home camera is just a “camera”. In addition to monitoring, there are not many additional functions, and the product value cannot be effectively released. The loading of AI can analyze and judge the behavior of the target through the camera, so that it can complete daily management and respond in time when abnormal conditions occur.

Through intelligent analysis, and with the help of face recognition, infrared sensing and other technologies, the home security products gradually move from simple security functions to consumer objects to the personal value satisfaction and sharing direction. In addition to security monitoring, it can also be Users provide services such as elderly care, infant care, and pet care. Communities, kindergartens, and nursing homes have become important application scenarios.

In addition, in the shops and other scenarios, real-time passenger flow statistics, advertising push, instant video sharing and other functions can be carried out to promote the transformation and upgrading of traditional stores.

"AI is the future development direction and it is unstoppable. Dahua Le Orange will fully participate in this project. In the future, the AI ​​of civil security products should be light enough, easy to use and cheap." Dahua Le Orange Business director Wang Yanbo told Lei Feng. In fact, the technology of the civil market and the industry market is open, but the form of technology display is different. Many industry technologies can be applied to the civilian market.

5. Evolution from "seeing clearly" to "understanding"

With the above-mentioned algorithm capabilities and chip performance further enhanced, in addition to the late intelligence, pre-judgment and analysis capabilities will become the next step in the security market.

Hu Mingshu, general manager of Dahua's stock business department, mentioned that the emergence of AI has greatly improved the efficiency of the traditional security industry and reshaped the business model of the industry. In the future, in addition to "seeing" the video surveillance cameras, security personnel are trying to make them "understand" the world. This also means that video surveillance is no longer a camera, but a person who can understand and think. Smart assistant.

Through the understanding of image intelligence, the camera can understand everyone in the video, each car, and then identify the person with the car, collect data; get the data, and then do the next data prediction.

Taking the face as an example, the face is now able to meet the most basic and core control services. In the next step, the application of the face technique will be carried out, and the face information collection, early warning and retrieval will be carried out, and the image-based early warning system will be used. Image analysis and travel law analysis are carried out for people within the scope of control. Full intelligence not only covers image intelligence AI, but also covers some core application scenarios.

In addition, human body-based video intelligence analysis can be performed on human postures and actions to generate richer business scenarios to change the operation mode of the entire police application.

The core part of the security industry is video surveillance. In the past, intelligent analysis applications were more reflected in simple alarms and snapshot information. What the user really wants is a perceptual intelligent analysis device, which is equivalent to one visual sensor in the Internet of Things, which recognizes the content in the monitoring picture, and describes it semantically and captures the best picture, and then passes the cloud computing. The platform analyzes and replaces people to think and judge.

Sixth, the security market competition is further intensified

In the past ten years, the competition in the security industry has been fierce.

After the price war and mergers and acquisitions in the security market, the market competition situation became clear, large enterprises rose rapidly, and the living space of small enterprises was further compressed. By 2018, the market "Matthew effect" will be further highlighted. The subordinates of Nuo's security industry may depend on the competition between several giants and rookies.

According to the "2016 Global CCTV and Video Surveillance Equipment Market Research Report" issued by IHS, in 2016, Hikvision and Dahua shares hold the top two video market share, and Yushi Technology ranks third, with market share being the first. More than 2 times the four.

However, the cruel "two-law" in the shopping mall allows people to remember only the top two in an industry. It is well known that Yushi Technology, which has a stake in it, has embraced Qianfang Technology on November 7 this year, introducing a comprehensive attack on capital. .

And this action may be the signal that the competition in 2018 is fully launched.

In addition, including AI soup, contempt, Yitu and other AI rookies, no accidents will also start to raise their sleeves next year, which means that the security market competition in addition to traditional security, "intrusion" It will also make this market even more beautiful. A security company executive revealed to Lei Feng. "The competition in the security industry has now reached the stage of pulling the network cable and shutting down the server switch."

And next year, this situation will only be even worse and will not be weakened.