Video Feature Compression Based On CDVS With Application To Surveillance

With the rising demand of the market and the progress of science and technology and the effect of other aspects, security monitoring market trend to have a booming development with intelligent and high definition. How to deal with the huge data generation from the increasing number of surveillance videos has been far beyond the ability of humans. The fast-growing of video data also result in a huge burden to data storage and data transmission through network. It is becoming a hot research topic to analyze?manage and retrieve the big data.Aiming at the problems of the traditional surveillance video analysis which is based on the Compress-then-Analyze(CTA) model, the problems including low performance of features, high storage space and high demand of transmission bandwidth, this paper uses the Analyze-then-Compress(ATC) model to deal with the videos. The CTA model, that is, the video compress with traditional video coding and transmission or storage, then analyze the compressed video. In this way, video analysis is based on the lossy content, which may result in low performance of video analysis and the increase of streaming transmit by network. The ATC model used in this paper mainly to applications of video analysis, sets of video features are extracted from raw videos then encoded before transmission, which required less torage space and transmission streams then the applications used CTA model. To extract the video features, this dissertation uses the advanced Compact Descriptors for Viusal Search to express the video static information and tracks the trajectories of key points using optical flow to express the dynamic information of video.In order to retrieve the surveillance video quickly, this paper builds a general framework of video retrieval system based on CDVS, then research the key technique of the systerm, complete the tasks including the video key-frame extraction, features of key-frame extraction and coding, tracing the trajectories of the key-point. For the technology of key frame extraction, the video sequences are divided according to the distance between the color histograms of the adjacent frames. For the technology of feature extraction, this paper analyzes the advantages and disadvantages of various features of video, and finally selects the CD VS features for video images, and analyzes the advantages of using CD VS feature points. Aiming at the key point trajectory tracking technology, this paper uses dense optical flow technology to track the key points in the key frame of the video, so as to analyze the video dynamic information with the trajectory of the key points.Finally,this work uses C++ and OpenCV to design and implement a surveillance video retrieval system which is based on CDVS.detailed system. The system has a good man-machine interface, experiment results showed that the system possesses advantages of real-time standard, higher reliability and precision and can meet the needs of surveillance video retrieval.