Research Of Content-based Surveillance Video Retrieval Algorithm

Video Surveillance is aimed at monitoring the action and movement of objects in sight by using cameras to capture the video image information of them. Nowadays, video surveillance has been widely used in various fields concerning the nation’s economy and the people’s livelihood, such as the security and defense, transportations and military, etc. Huge amounts of video files are produced by the video surveillance system, which are huge in terms of their quantity with complex structure and various forms. As a result, it is hardly possible to meet the requirements of real work in terms of the time limit and accuracy if people employ the traditional text-based markup methods to browsing and retrieval the surveillance video. To solve this problem, the thesis studies an efficient way to look through and retrieve video, namely-esearch of content-based surveillance video retrieval algorithm.Researching the basic theory of content-based video retrieval algorithm, and considering the features of video surveillance scene and people’s need for retrieval, this thesis focuses on the study of the key algorithm of video retrieval, including shot segmentation, key frame extraction, key frame image matching and retrieval. The following points are to be addressed:(1)Thoroughly analyzing the applied values and perspective of the content-based video retrieval algorithm; Looking at the development of this field both home and abroad; Studying and analyzing the basic theory and some common retrieval algorithm of it.(2)Doing research on and putting forward a shot segmentation algorithm based on grayscale change detection in light of the features of video surveillance scenes and the practical needs. By setting the virtual detection line, Statistical and computing grayscale change lines of virtual path to determine the start of the lens; calculating the target gray value of the overall prospect when it is reduced to a certain value marked the end of the lens. Then the video shot is captured.(3) As for the extraction of the Key frame, decide the first key frame rationally and then update and obtain the remaining key frames by using edge features to calculate the difference between frames.(4) Studying a retrieval matching algorithm of the key frame and putting forward an image retrieval algorithm based on edge direction histogram correlation matching. Figuring out the edge direction histogram after denoising the images and extracting the edge. Forming the feature vector through ranking the histogram. Working out the correlation coefficient between the figure eigenvectors by using Spearman rank correlation formula, which can be used as the criteria to measure the similarities between figures. The validity and reliability of this algorithm is to be tested through experiments.(5) Doing research on how to use the features of color and shape to match key frame in a comprehensive way in the light of the fact that people often pay attention to the colors and shapes of the objects in a surveillance video.Finally, Combing with the website development technical and the algorithm we study in this paper, an online surveillance video retrieval system was developed and users can remote login the system and retrieval the video.