Research On Processing Multi-Streams In Intelligent Video Surveillance

Intelligent video surveillance (IVS) is one of the branches of computer vision. Methods in computer vision and image processing are used by IVS to deal with image sequence. These processes include detect moving object, classify moving object, trace moving object. So the behavior of moving object can be understood in the scene. IVS can be used in production, security system, traffic surveillance and so on.Monitoring system can display many videos by one video server, but it can't deal with them intelligently. Substantive research of IVS is just on one video stream. So the traditional intelligent video surveillance technology must be changed to achieve better practical application. In the paper, the problem how to deal with multiple video streams was studied with the view of Image Engineering. And the breakthrough point of this problem is how to increase the speed and ability of algorithms. Through previous research, this paper adds some new method to deal with multi-streams.Firstly, in the paper, the classic techniques of IVS are reviewed to improve IVS, including some widely used algorithms and discussed the advantages and disadvantages of these techniques.Secondly, I make some improves on the first step in video processing and design a new image pre-processing method. This method can change its structure dynamically based on two rules. One is gray-area rules; the other is local differences rules. So the method has the advantages both of median filter and wavelet filter. It is certified fit to deal with multi-streams.Secondly, because the processing of multi-streams has it own requirement and specific application environment is different, we proposed two new methods to detect moving object. One method simulates the human memory process. This method let both current video streaming information and history video streaming information to make contribution in detecting. This method was proved that it has low computational complexity and better than the finite-difference time-domain method. The other method is used in production and base on static background-data library. Combining database technology and detect-object technology which appear in this method will be a new way for IVS. The method was proved that it is particularly suitable for surveillance production and can deal with the change of the environment well and can surveillance effectively even in more extreme cases. After that a problem is discussed that the IVS will miss some object information. The problem has be paid less attention by now, while it has a great impact on the subsequent treatment. To save this problem effectively, some classical methods and methods are used.At last, relatively stable visual features of object are needed in order to achieve the goal of cross-identification in the multi-streams. So some new methods are brought about cross-identification. Some of those methods come from video-search which is one of branches of computer vision too. This paper presents a new visual feature-contour variability. This new visual feature can make the information of the object movement visually. An algorithm is designed to quickly identify at the some time. The algorithm uses mask technology and base on the serial integration of multiple features. It was proved that this algorithm can increased the speed of dealing with multi-streams.IVS can monitor more than one scene by doing these works in the paper. I used the new IVS to monitor the production of steel pipe in Pan Gang Group Company LTD. The result is that these works are effectively.