Giants and AI rookie snatch the security market, landing to cross these thresholds
Intelligent monitoring is used to analyze massive data in real time to extract effective clues, so as to lock the trajectory of suspects or vehicles, complete tracking and capture, or face recognition access control system in office buildings. The outbreak of AI makes security more and more close to daily life.
Whether it is haikangweishi, Dahua shares, Yushi technology and other established giants in the security industry, or new AI companies such as Shangtang Technology, Kuangshi technology and Shenjian technology, AI security is the first gold mine, competition for food 600 billion market space.
For novice opportunities, traditional security is a highly centralized industry, with the market share of the three major security giants accounting for more than 50%.
However, because artificial intelligence technology can quickly use video structured description technology to extract structured data from video images, this brings AI new companies the opportunity to tear apart the market gap.
However, this is not an easy market to break through.
For traditional security giants, the algorithm advantage of AI start-up companies cannot easily take a closed loop of security industry chain, which is different from other markets, the company must have a thorough understanding of customer needs, the entire sales system, product system and solutions, and technology must be integrated into the front and back ends of the entire industry chain to gain a firm foothold.
For example, in the security market, government business is the most important, and various governments have different procurement standards and requirements for security products, and the region is very obvious.
As a highly closed market, purchasers are also more inclined to choose manufacturers with strength and rich experience in the security industry.
Judging from the bidding process and standards, the qualification and certification materials of suppliers are strictly required, which to some extent keeps many new AI manufacturers out of the door.
At present, AI companies entering the security market are mainly divided into two categories. One is image recognition technology companies that provide software solutions, such as Shangtang Technology and Kuangshi technology;
The other is chip companies that provide hardware solutions, such as Cambrian and horizon.
For these companies, they often have advanced algorithms, but they lack profound industry accumulation and industry resources, thus taking many detours.
At the beginning of entering the security market, Kuangshi technology once poached people from giant companies such as haikangweishi and Dahua shares and set up branches in Hangzhou, but it was difficult for the team to integrate due to regional differences and technical background problems.
This forced them to go deep into the market and create demonstration models in cities.
The security industry has a real, rigid and large-scale demand for AI technology, and the technical difficulty is also extremely high. For example, Shenzhen's face recognition project connects tens of thousands of videos to the same system, you need to search in hundreds of billions of photos.
This not only requires extremely high face recognition accuracy, but also is very important to the speed of face capture and recognition. At the same time, the chip should be fast, small and good, this has very high requirements for algorithm innovation, algorithm optimization, network structure pruning ability, etc. '
John Young, co-founder and vice president of Shangtang Technology, for example, security is the fastest growing and largest market in the field of artificial intelligence, but the overall proportion of real effective intelligence is still very low. '
Although the algorithm will have the first-mover advantage for a period of time, with the mature application of GPU and depth algorithm, these advantages are easily weakened, companies, including Shang Tang technology, have chosen to cooperate with traditional companies to get a share, go deep into the vertical industry and avoid positive competition.
In terms of capital selection, these newly entered companies are also paying more and more attention to the blessing of resources, especially investment funds with government background, which are more attractive than simple financial investment.
The front-end intelligence and the power of start-up companies have also forced Giants to step up research and development of AI and cutting-edge technologies.
Because security is a 2B business in public security applications, the emergence of AI has changed our entire business process. After the front-end information collection, it is no longer just video, there are also pictures and structured information, from storage to background analysis, which is a whole process challenge. '
Yao Hua, vice president and chief architect of Yushi technology, said.
Although security is the hottest application of AI, the technical difficulty is not low.
One of the problems is that artificial intelligence has poor adaptability to the recognition environment of video content, and is easily affected by environmental changes such as lighting conditions, weather factors, image quality, target size, ground object occlusion, etc.
Complex light, fast movement and side face occlusion are all deadly enemies of front-end face recognition. 'Yao Hua said.
In addition, a large number of videos, images, structured semi-structured data writing and reading pose challenges to storage.
In Yao Hua's view, when the structured read-write ratio of AI data becomes 1: 1,7 X
During 24-hour real-time writing, it will bring great challenges to the entire hardware system, and many high-end storage will fail.
At the same time, the new data generated after high definition requires higher reliability. How to store it reliably with lower cost and generate higher value is the current problem.
Based on the average cost of current storage devices accounting for 8% of video surveillance systems, this is indeed a big expense.
Therefore, front-end intelligence is regarded as one of the future trends of security.
In the past, we used video coding to compress, some of which were lossy compression, while front-end sensors could obtain full-size data and process it directly on the original data, thus optimizing the data quality. '
Xiao Hongbo, CEO of touch view unlimited company, which builds front-end embedded security AI based on AI chips, stressed.
From the perspective of security and privacy, devices do not need to be connected to the internet and are processed locally, which can also alleviate hacker intrusion, lack of security access control and pose challenges to information security to a certain extent.