The difference between intelligent video analysis and traditional industry products is that its core algorithm is still far from being finalized, and it has a lot of room for expansion, because with the popularization of applications, the functions of products will be more specific Claim.

The demand for intelligent video analysis is an inevitable outcome after the level of video surveillance has reached a certain level. It can well alleviate the problem of massive information redundancy in traditional video surveillance. Although after several years of development, at the present stage, both the technology and application of intelligent video analysis and the market environment are still in a stage of growth, and they need to be continuously explored in practice.

Obvious differences in demand in different industries

       It is understood that the demand for video surveillance in different industries generally has very obvious differences, especially for the application requirements of intelligent video analytics technology, which also determines the specificity of detection behavior types and abnormal events among different industries. For example, in a safe city, it is possible to develop behavior analysis functions such as fighting, robbery, and climbing in response to public security incidents in the city; in the bank's ATM self-service area, it is possible to achieve illegality by analyzing the current criminal means of the ATM self-service area. The function of analyzing behaviors such as sticking small paper strips, installing fake keyboards, masking, violent robberies, etc. In the transportation industry, more functions can be implemented to alert traffic incidents such as retrograde, illegal parking, and traffic congestion.

Only by combining the practical applications of the industry and in-depth understanding of the specific issues of different industries, can we better grasp the needs of users and put the functions of intelligent video analysis technology into practice. This is also the ultimate reflection of the future industrial value of intelligent video analysis technology. .

Constantly improve the core algorithm

       In terms of technology, current intelligent video analysis technologies are mainly divided into pattern recognition and behavior analysis technologies. Pattern recognition technology includes: face detection, face comparison, license plate recognition, pyrotechnic identification detection, etc.; behavior analysis includes: crossing lines, entering, leaving, missing, leftovers, crowd gathering, slipping and so on. The core of intelligent analysis is the video algorithm. Currently, the embodiment of the algorithm includes the integration of hardware and software, and the results are basically the same.

At present, there are more and more manufacturers focusing on the field of intelligent video analysis. Comprehensive enterprise representatives such as Hikvision, special intelligent industries such as Zhuo Yang Technology, Zhi An Bang Technology, Wen An Technology, etc. U.S. ObjectVideo, Israel's ioimage, NICE, etc.

       In order to do a good job in the promotion and in-depth application of intelligent analysis, from the aspect of manufacturers, it is still necessary to continuously improve the core algorithm technology of the product, or to develop more application modes that conform to each specific scenario. The difference between intelligent video analysis and traditional industry products is that its core algorithm is still far from being finalized, and it has a lot of room for expansion, because with the popularization of applications, the functions of products will be more specific Claim. In addition, the technology of intelligent video analysis often involves many complex parameter configurations. How to simplify the configuration of these technical parameters is a problem that manufacturers need to pay attention to. At the same time, the promotion of product knowledge and industry technology involved in the process of industry promotion also requires the joint efforts of various manufacturers.

       At present, the actual complex application environment requires more sophisticated algorithms. Currently, the application of intelligent video analysis technology to develop embedded products with corresponding advanced functions solves the difficulty of construction and the resistance of practical promotion of many products. This is also the development of intelligent video analysis technology. The inevitable trend.

Reduce missed reports and false alarms

       An intelligent video analysis product that uses the detection accuracy as an evaluation criterion generally evaluates its performance by examining its detection rate and false alarm rate in different environments. The detection rate and efficiency of the ideal system should be 100%, but the actual system can only be close to these indicators rather than fully achieved. Only by adjusting the sensitivity parameters of the system and changing the relationship between the detection rate and the efficiency, the system with excellent performance can achieve the highest detection rate when the efficiency meets the requirements of the use. When the sensitivity is adjusted high, the detection rate increases but false positives increase and the efficiency decreases. On the contrary, the false alarm rate is reduced, and the efficiency is improved but the detection rate is reduced.

       Intelligent analysis can achieve simple behavior analysis. Improve the form of false negatives and false negatives must be designed in a holistic way, such as the location of the camera, the selection of the lens, the light in the room, the filter and sensitivity, and the reasonable setting of the rules. Through overall considerations, the false alarm rate can be reduced.

       The effects of smart video analysis products on the market today are uneven, and some smart flags do not actually achieve smart results. In the future, the manufacturers will continue to improve the algorithm, and at the same time strengthen the testing in complex and harsh environments, continuously improve the performance of the product, and minimize missed reports and false positives.

The part that needs to be improved in the future

       Although intelligent video analysis technology has been rapidly developed, there are many complex technical problems in implementing it into specific applications.

       Can not completely eliminate the impact of false positives. For example, background modeling technology in moving object recognition can not completely eliminate false alarms while controlling the number of missed reports. In the perimeter prevention products with target recognition as the basis of technology, the number of false positives has always been a major indicator reflecting the pros and cons of the product. The number of false positives is caused by the difference between the background model and actual use. The stronger the model's adaptability, the less false positives are caused and the higher the required technology is.

       There are many factors that affect the establishment of the background model, such as the empty asphalt road and the wall with trees on the edge. The images taken by a 6m-high camera and a 2m-high camera require different background models. The models needed for daytime and nighttime conditions also need models. different. At present, the industry has not yet developed a background model that can cover all use cases, nor can it completely solve the effects of random events, such as false alarms caused by lights in the darkness.

Do not have the ability to judge behavior. For example, perimeter prevention, the regional intrusion function of behavior analysis in the machine can find the activity target, and can use the technical means to extract the target (such as the human body) that the user wants in these activities. But further, what is the motivation of this intruder, whether it is an occasional pass or deliberate intrusion, whether it intends to visit the guarded area, which cannot be identified by the machine. After all, intelligent video analysis is still just a series of well-established mathematical formulas and procedures that are far from human judgment.

       Feature recognition technology has high requirements on the picture. For feature-based analysis techniques, the requirements for images are relatively high. In addition to the definition of the screen itself, it is also necessary to clearly display the characteristics of the target object. Currently, the recognition ability of the computer is much lower than the ability of human to recognize the characteristics of the object. Different lighting conditions and shooting angles will change the characteristics seen by the computer. . Therefore, the requirements for the installation of the camera and the surrounding environment are relatively high. For example, license plate recognition products have relatively strict requirements on the angles and pixel sizes presented by license plates on the screen. These high requirements limit the implementation and application of such products.

       The processing power of the CPU is still a bottleneck. To meet the actual complex application environment, more complex algorithms are needed, which brings a huge amount of calculations. Currently, the capabilities of DSP chips are limited, and they cannot meet the needs of some complex algorithms. Therefore, it is difficult to develop embedded products with corresponding advanced functions, which also increases the difficulty of construction and the resistance of practical promotion of many products.

       In general, intelligent video analysis technology is also like a computer. It is smart for traditional means, but it is still at a low level for human intelligence.