Research On Intelligent Video Surveillance Technology

Intelligent Video Surveillance technology is an emerging research orientation in the field of computer vision. Its main goal is to realize the description, understanding and analysis of the content of the surveillance video by integrating computer vision technology, image/video processing technology and artificial intelligence technology. The analysis results can be used to control the video surveillance system itself. Thus, the video surveillance system will be improved to a higher level. The research contents of intelligent video surveillance include: moving object detection, object representation, object tracking, object recognition and object behavior analysis. Their research results can be applied into public safety protection, medical care, traffic management, customer service, and many other fields. Therefore, intelligent video surveillance has a profound theoretical value and broad application prospects.This paper is committed to the key issues of intelligent video surveillance technology. And the research work mainly covers the following topics:In the area of the description of video moving objects, two new shape features: a boundary descriptor based on directional distance vector and a wavelet boundary descriptor are presented to improve the efficiency of exiting methods. The latter is more excellent. This descriptor can represent shape very accurately no matter whether the shape is convex or not. Furthermore, it is invariant to shape translation, scaling and rotation. In addition, it can be used to retrieval shape under multi-resolution. Therefore, this descriptor is very important for the realizations of object tracking, recognition and behavior analysis.In the area of object tracking, firstly, a multi-object tracking method based on linear prediction and shape matching is presented. The tracking result of this method is very ideal no matter whether the object is rigid or non-rigid body. The obtained curve from the tracked object is agreeable to the its true boundary. So, this curve can be used to object recognition and behavior analysis. Furthermore, the stability of this tracking process is excellent. Even if the speed of moving object changes sharply, it is still can be tracked exactly. In addition, the computational complexity is low enough to make sure the real-time object tracking. Secondly, a multi-object tracking method based on mean-shift and shape matching is presented. This method is improved from the last method. The linear prediction in the last method is replaced by the combination of mean-shift and linear prediction in the new methods. This improvement make the tracking more robust.In the area of object recognition, firstly, a segmentation method based on mean-shift and region growing is presented. Secondly, a image compression method based on Jacket matrix is presented. Finally, a new recognition method based on wavelet boundary descriptor and support vector machine is presented. The first two methods are applied in image preprocessing. They are useful to construct training sets for object recognition. The last one is the most important method for object recognition. Its main idea is to distinguish different video objects according to their shapes. The feature used here to represent shape is the wavelet boundary descriptor mentioned above. With this descriptor, a new recognition framework based on support vector machine is constructed. This method can distinguish not only human, animal, vehicle, but also human's many kinds of postures, such as: standing, sitting, and lying and so on. So this method provides behavior analysis with a powerful tool.In the area of behavior analysis, a three-layer model method is presented. This method can be not only used to analyze the behaviors of individual object, such as: running, jumping, creeping, but also used to analyze the cooperative behaviors of multiple objects, such as: object liens, crowd gathering. Furthermore, the analysis of the various behaviors are independent of each other. In addition, the calculating speed of this method can satisfy the requirement of real-time video surveillance.Finally, an image retrieval prototype system and a human behavior analysis prototype system are built in order to improve and validate all kinds methods.