Video analysis and identification (video analyzing and recognition) technology refers to the use of computer from the video through computing and analysis, extract the useful information in the video of a technology, which is to extract the video of "content" and understanding. As if the people see a period of "one car" in the video, "there is a white car", "there is a white jetta car", "there is a white jetta car going on right turn signal to turn right". For people, this period of video is meaningful, is contains a certain amount of information, and people can intelligently extract this information, access to "have a white jetta car is preparing to turn right and right turn signal" this information, and this information contains a "car, white, jetta, lights, lights, a right turn signal, ready to turn right" the series of information. While video analysis and recognition technology is to make the computer to complete the process of information extraction and understanding, often can also be referred to as "video analysis technology.

It must be noted is two things: one is "video analysis technology" is also sometimes called "video image analysis technology," why? This is because the video itself is composed of a series of continuous images (don't discuss the video compression technology, here only refers to the video signal through the decompression after restore image sequence of frames), the understanding of the video content, is based on the "image sequence analysis and recognition, therefore, the two are equivalent, is the meaning of the same; Second, most of the time the information in the video is very rich, as mentioned above this piece of video, in addition to this white jetta car, there may be additional information, such as "standing on the side of the road is a middle-aged man wore a dark blue coat wearing sunglasses, smoking a cigarette", for the same period of video, we focus on the object is different, need to extract the information of different, the human brain can handle very complex work, can at the same time to pick up the most of the video information at once, and for the computer, the intelligent level is relatively low, perhaps only targeted to extract some information, such as information or just take out the car only to extract information. But what kind of information extraction, belong to a kind of "video analysis technology.

Video analysis technology

A wide range of video analysis technology, the front said, as long as the operation to extract useful information from video can be called a video analysis technology, because are all belong to the "video content analysis, recognition and understanding, from this perspective, is now more mature and has formed the product applied in the actual project technology belong to the video analysis technique, such as license plate recognition technology, video retrieval technology, video face detection, etc., because it belongs to the extraction of useful information in video, to extract the license plate number, the extraction of video text or graphics, etc. Strictly speaking, these techniques are of video analysis technology is simple, just because of these technologies have been well research and application, there are some special terms, to carry out alone, and seems to be no longer be generalized to the category of "video analysis technology".

Because of habits, the current video analysis technology generally refers to the behavior of target motion analysis from video, extraction and recognition. It means than the literal meaning of the meaning of meaning has narrowed considerably.

Technology research direction

Current for video analysis technology (especially after the narrow definition of "the behavior of target motion analysis, extraction and identification of" the concept), in general is divided into two main research direction: a research direction is the target of the whole trajectory as the research target, to extract the moving object characteristics of the movement or inherent in its characteristics. This research goal is not necessarily mean the people, but can be in any moving object as the research object, such as people, vehicles, animals, aircraft, tanks and other military targets and so on. Refers to is the research object and the behavior characteristics of the movement, such as whether the target movement in a particular direction, whether in a particular trajectory, whether a given alert across a line or to enter areas, whether queuing, whether produced gathered or congestion, whether there are following phenomena, whether there is a wandering behavior and so on. This is an important research direction, and have corresponding products appear on the market (although is not very mature). Its characteristic is the target object as a whole to examine, extract the movement characteristics of within the scope of the big scene.

Another research direction is to target the local part of the movement as the research target, to extract the local "body language" features, such as video sign language recognition, gait recognition, face recognition, or whether it upon completion of a certain behavior, such as a phone call, put down a package, from somewhere or into a thing, and so on. In addition, a public place or whether the scope of cultural relics with cadence, the movement posture or movement in the sports action is the best, even by shipboard gun barrel rotary motion curve of the research, all of these are belong to the same direction. The research direction, which is usually aimed at small view scene close shot video are studied, and its research object is only belong to the local part of the target, such as analysis of man's hands, feet, head of the movement, the most core implementation steps usually includes 3 d modeling of object of study. This research direction due to the analysis of the action is more detailed and specific, so that most belong to that kind of for the development and application of a specific demand, is more difficult to general modestly higher forming products.

Market development

There is no denying the fact that, for video analysis technology research, but as a result of its algorithm complexity and the multiplicity of the target behavior, development has been slow. And relatively, as a result of the need of international anti-terrorism situation, and the first kind of research can be used in the monitoring system realize the function of the abnormal state automatic alarm, abnormal events with real-time alarm, shorten the reaction time, reduce the loss, enhance deterrent monitoring and control system, etc. Abroad in the field of video analysis technology research, development and transition of domestic research institutes, research results have certain leading.

Although according to different application requirements, the products have different target market and target users, its function also is not the same, but many of the core technology in the process of its implementation and implementation approach is the same, all need to solve some common problems, and then according to specific needs to add some special processing and operation, higher precision, faster and higher accuracy.

Research direction in the first category, with the goal of the whole trajectory of the extraction and analysis, for example, although its processing techniques or methods are different, have distinguishing feature each, but from the overall solution framework design, is to get to background, and then extract the prospect goal, reanalysis prospect target trajectory curve, finally by the characteristics of path curves to realize the analysis of abnormal behavior. In the process, inevitably involves is for light, shadow and abnormal state such as jitter, fuzzy and adapt to, to adapt to the outdoor or indoor illumination change bulb strength change; Adapt to all kinds of shadows, including the shadow and the shadow of the target itself; Adapt the camera dithering and leaves, water wave, the refraction of light caused by the jitter; Is not allowed to adapt to the focal length or sleet fog caused by video fuzzy, etc. Under the condition of these unfavorable external environment will still be able to accurately extract the target trajectory and analyzing the behavior, can accurately report to the police as much as possible and reduce the false alarm false alarm, to ensure the effectiveness of the automatic monitoring. In this at the same time, must try to consider the speed of operation, the complexity of the algorithm, to ensure the timeliness of alarm. Only in this way, could the development of video analysis technology into products can be applied.

In fact, currently on the market already appeared the intelligent video analysis technology of products is still in a very low-level stage, can only analyze abnormal behavior of several relatively simple, its light environment adaptability index is in a lower level, so to speak, is still relatively low, the degree of "intelligent" away from the ideal effect of the user's expectations are still relatively far worse, but is not to say that these products can't use, still can be used, the question is how to use, how to use.

Product performance

Then, in the present is still in the comparison "primary" should be how to use the intelligent level of video analysis technology product to perform its functions? From three aspects to consider:    One: the application of customized product, the custom here is not necessarily the pointer to each application specially developed a set of algorithm (although to do so in theory should be the best, but impractical), but for a specific application scenario and application of the target, should set some parameters as much as possible, including the rule parameters and even internal parameters of the algorithm, the algorithm can get the best performance in the environment. With strong sex, to which can product is not good, just for the moment does not appear yet, if the relative specific application targeted set specific parameters, the performance will be greatly improved.

Second: the application in a special occasion. As for the target trajectory monitoring and alarm video analysis technology products, is one of the key link for the extraction of background. When visual field goals (foreground) is large, to extract the background, there will be a larger error is not conducive to the alarm accuracy, you can choose to apply prospect target less occasion. As this product in the tiananmen square this crowded environment is certainly effect is poorer, but if used in the military restricted zone or the bank vault door, under normal circumstances is very moving targets, so it will effect a lot better, by the same token, the effect is also used in the urban road monitoring must use on the highway.

Third: can be used in place of alarm accuracy requirement is not high. If some applications require that the alarm accuracy is very high, once appear, false alarm or leakage alarm can cause very serious consequences, so for the performance of the product is very strict, this is not quite suitable for the application of this kind of intelligent level is not high. Only when using only need to automatic monitoring of auxiliary functions, allowing a misstatement or omission of, can be used to the existing video analysis technology products.