China's Security Intelligence Status

1. Smart security applause: The ideal fullness, the realistic sensibility Although it is the "pampering" technology in the market, the application of intelligent video analysis technology in China is not easy. Some of the world's professional IVS research manufacturers, such as the United States, ObjectVideo, Vidient; *** NICE, Mate, IOImage; Australia's IOmniscient and so on. These manufacturers have entered the Chinese market one after another, causing foreign companies to monopolize the domestic market of intelligent video analysis. In the end, because of the difference between its technology and China's national conditions, it cannot really penetrate the Chinese market. However, most of the domestic smart video analytics manufacturers can identify the behavioral features that are limited to specific situations. For example, target tracking, cross-border, counting, target loss, object left, etc., are mainly used in relatively fixed application scenarios such as detention centers, prisons, museums, warehouses, factory areas, and subway stations, and in monitoring environments targeting the social level, due to monitoring. The complexity and variability of the scenarios have led to a significant reduction in the stability of the monitoring. The intelligent monitoring products that use the accuracy of the alarm events as the evaluation criteria have also highlighted their application limitations.

Intelligent video analytics technology is in the embarrassing position of “what you can call it or not”. If we investigate where the inflection point of intelligentized large-scale application is, it should be said that in addition to the innate advantages of the previous analysis of its own technology, it is also necessary to address the following: Bottleneck problems, perhaps, sand and sand towers, the explosion of security intelligence is just around the corner.

2. Technology can hardly meet application requirements At present, although smart video technology has developed rapidly in recent years, the application range is also expanding. There are various products for network monitoring on the market, such as network cameras, network matrix, etc. The quality of the system has yet to be improved, and the images appear to have obvious delays, jitters, and lack of clarity. Because the hardware itself is not stable enough, problems such as crashes, restarts, and missed reports may occur. The limitations of the intelligent video analysis algorithm itself can realize all its functions in an ideal environment. Therefore, in practical applications, only simpler functions can be implemented initially.

Can not completely eliminate the impact of false positives For example, the background modeling technology in moving target recognition, while controlling the number of missed reports, can not completely remove false positives. 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.

Another example is the empty asphalt road and the wall with trees on the edge. The images taken by a 6-meter-high camera and a 2-meter-high camera require different background models. The day and night conditions require different models. 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 dark environment; secondly, it does not have the ability to judge behaviors. . For example, the perimeter intrusion prevention and the regional intrusion function of behavior analysis in the machine can find the target of the activity, and can use the technical means to extract the target (such as the human body) desired by the user. But further, what is the motive of the intruder, whether it is an occasional pass or a deliberate intrusion, whether it intends to visit the guarded area, which cannot be identified by machines?

Feature recognition technology has high requirements for images. 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. The current computer's recognition ability is much lower than human's ability to recognize the object's features. Different lighting conditions and shooting angles will change the characteristics seen by the computer. Finally, 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.