Intelligent Analysis Application and Development Trend

The domestic video analysis technology has also been rapidly developed, but domestic smart analysis applications should start from 2005, from the early application of behavior analysis to the current deep application of various industries. With the rapid expansion and development of the security industry, domestic video analytics has developed very rapidly.

The intelligent video analysis technology (Video Analytics) integrates multidisciplinary research results, mainly including image processing, tracking technology, pattern recognition, software engineering, digital signal processing (DSP) and other fields. Since the September 11, 2001 incident, the United States has increased its investment in video analytics technology. The data shows that from 2002 to 2005, it was a peak period for video analytics technology research. The current state of application of video analysis technology From a technical point of view, the current domestic intelligent analysis technology is mainly concentrated in two major categories: One is the use of screen segmentation foreground extraction method to extract the target of the video screen, through a variety of different Rules to distinguish between different events, so as to achieve different judgments and generate corresponding alarm linkage, for example: some of the earliest behavior analysis functions (cross-border, regional invasion, fight detection, personnel gathering, etc.), as well as early traffic incidents Detection and so on all belong to the application of such algorithmic techniques. The other is the use of pattern recognition technology to model specific objects in the picture and train through a large number of samples to achieve detection and application of specific objects in the video picture. Such as vehicle detection, face detection, head detection (flow statistics) and other applications.

From an application perspective, there are currently four major categories of domestic intelligent analysis technologies: The first category is real-time alarms. The second category is the data statistics category, and the third category is the property identification category. The fourth category is image processing.

The first type, real-time alarm class, mainly analyzes and judges the content of real-time video through the analysis technology. When a certain state reaches the requirements of the alarm rule, the system can send alarm linkage. Such as the most basic cross-line alarms, intrusion alarms, fight alarms, crowd alarms, etc. Of course, with the deepening of applications, there are many real-time alarm applications with industry characteristics in various industrial applications, such as the traffic industry's congestion alarm, Pedestrians on the high-speed alarm; the judicial industry's climbing alarm, leaving the alarm; the financial industry's trailing alarm, fake advertising quotes and so on.

The second category, data statistics. It is mainly through the statistics of specific content in the video content under specific scenarios to form relevant reports and data applications. For example, through the video analysis, the auto traffic statistics on the highway are calculated, and the traffic statistics such as the entrances and exits of malls are statistically analyzed through video analysis.

The third category, attribute recognition class, is to automatically identify the properties of a particular transaction in the video to achieve in-depth application of video content and fast retrieval. Such as face recognition, license plate recognition, car logo recognition, color recognition, gender identification, height recognition, age recognition, gesture recognition, and so on. At present, most of the applications in the security industry are face recognition, license plate recognition, and vehicle logo recognition.

The fourth category is image processing. Mainly for the image as a whole analysis and judgment and optimization to achieve better results or to clear the contents of the algorithm through the calculation and processing to achieve a clear effect. Such as current video enhancement techniques (denoising, defogging, sharpening, highlighting, etc.), video restoration techniques (deblurring, distortion correction, etc.).

In terms of product form, currently there are two main product types in the market, one is front-end smart products. One is the back-end server product. These two types of products have their own advantages and disadvantages, depending on the application and type of project.

At present, many smart cameras can be seen on the market, that is, some video analysis algorithms are transplanted to network cameras, real-time video analysis and inspection are implemented in the cameras, and some intelligent analysis functions are implemented (such as vehicle capture cameras that currently implement vehicle detection. (Cameras for detecting alarms that achieve regional intrusion and other functions, cameras that implement human detection and capture, etc.), smart cameras have their own analysis capabilities, and the system architecture is simple. At the same time, the current camera DSP processing capabilities are fully capable of running a variety of complex functions. The analysis algorithm has the same analysis results as the back-end products.

The intelligent analysis server should be still a mainstream product in the market. The server products have the characteristics of short software development cycle, flexible project application, and strong adaptability of the transformation project. At the same time, there are still relatively complex intelligent analysis functions that require a large amount of calculations. Complete porting to a camera requires substantial optimization and improvement of algorithm performance and increases the cost of front-end equipment hardware. The x86-based server is the best choice for running these complex algorithms, the cost is relatively low, and the best analytical performance can be achieved.

For example, most applications such as current passenger flow statistics analysis, face search and analysis, face comparison, traffic incident detection, video content retrieval, and video quality diagnosis are still concentrated on server products. There are fewer, if any, transplants to cameras. Portions of less computational content are ported to the camera for processing. But server products have an indisputable disadvantage, that is stability. Stability is certainly not good for embedded products. Therefore, many manufacturers will still try to optimize the algorithm to transplant various functions into embedded devices as far as possible under the premise of cost control, so as to improve the stability of the system, especially some devices that need to be used in outdoor environments (such as traffic Event detection and other applications).

There are big problems in the application of intelligent analysis technology. Early video analysis application products appeared on the market, which really caused a commotion, and many special application scenarios and application environments can indeed bring great value to customers. Such as the mall's flow statistics technology, for the mall's data analysis has brought huge technical support. Such as license plate recognition technology, the value brought to public traffic management cannot be measured by data. However, video analytics technology has not yet fully matured. At present, it should still belong to the initial stage of technology application. There are still many problems. These problems may also be the most important factors that limit the rapid development of video intelligent analysis applications:

(a) The accuracy of detection does not achieve the desired results. The accuracy of video analysis technology can hardly achieve very satisfactory results. In particular, the application of real-time alarms, false positives and false negatives are all of the most concern to customers. If the false positives are too high, customers can't stand it. Missing reports, customers can not stand more. In particular, some of the applications requiring higher levels will not be effective as long as they are leaky.

(B), subject to large environmental disturbances. One of the biggest problems in video analytics technology is that interference from the environment and video quality is too great. Light, sundries, bad weather, sloshing, flying insects, etc. will interfere with the application system, and the application system will be very poor or even fail to work properly. .

(III) Installation and debugging are complicated. Intelligent analysis application products almost all need to carry out different parameter debugging according to each application scenario, and it involves a lot of professional parameter debugging. Non-professionals simply cannot debug the desired results.

The development trend of the application of intelligent analysis technology Overall, the biggest factor limiting the application of intelligent analysis technology is the accuracy rate. So the development trend of the application of intelligent analysis technology is definitely moving in the direction of improving the accuracy rate. At the same time, on the other hand, everyone will also look for some application directions that do not care about accuracy and pay more attention to efficiency.

There should mainly be several developments:

(1) Increase the judgment information from the source. The promotion of the binocular camera should be a general direction. The binocular camera has two shots and the acquired video has depth information of the target. The analysis algorithm calculation can follow this information and accurately determine the distance, depth, height and other information between objects, which can improve the accuracy of the overall algorithm.

(B), a variety of self-learning and adaptive algorithm research and application. Subsequent smart analysis products should have strong self-learning and adaptive capabilities. It can automatically learn and filter according to different complex environments, and can automatically filter some interference targets in the video. In order to achieve improved accuracy and reduce the complexity of debugging purposes. For example, anti-dithering algorithms, repeated moving object filtering, automatic filtering of tiny objects, automatic light suppression, 3D modeling, and other technology development and in-depth application.

(C), the rapid development of video data mining applications. With the rapid development of video analysis technology, the amount of video data is also very large. How to make video analysis technology play a role in big data has become a focus of attention. Using various algorithms to calculate, a large number of things in different properties of video data are used for searching, labeling, identification, and other applications to achieve fast search and retrieval of large amounts of data. Greatly reduce labor costs and increase efficiency. There are even aspects that make it impossible for some human to complete the task. Such as: large face database search, identity card library duplicate staff search, wear a certain kind of clothes in the video, a certain color of vehicle search, license plate search, and even can do map search applications (Enter a picture to find and pictures Similar fragments).