Research On Key Techniques For Content-based Surveillance Video Retrieval

In recent years, people pay more attention to the importance of security. As a result, more and more intelligent video surveillance systems have been applied to the bank, school, community and other security-sensitive situations. However, the ability to analyze and process the surveillance video is still relatively limited, especially in the face of the current massive surveillance video database. People hope to propose a technique through which they can quickly and efficiently find what they are interested in. Content-based surveillance video retrieval technique is one of such kind of solutions, which make use of computer to analyze the content of surveillance video automatically, extracting feature into a index. When users submit retrieval needs, the retrieval system will automatically returned the images or video clips closest to the retrieval needs.This paper analyzes the key issues in intelligent video surveillance system such as the moving object detection, video segmentation, key frame selection, similarity measure and so on, and also brings out some reasonable solutions to these issues. First, a code-based algorithm is utilized to detect moving object on the basis of the test results of frame-difference method and mixture of Gaussians method. Second, an automatic video segmentation method by counting the movement of the current frame is proposed mainly using the foreground and background information in moving object detection. Finally, a content-based image retrieval method with color and texture features and a video clips retrieval method with SIFT features are also proposed and realized. Based on the above algorithms, a content-based surveillance video retrieval system, named CSVRS(Content-based Surveillance Video Retrieval System) is designed and implemented. And lots experimental results have proved the validity and effectiveness of the algorithms.