Research On Some Key Technologies Of IP-based Video Surveillance

Following the development of the technology of electronic, information, computer and network, the technology level and actural usage of video surveillance system has achieved much progress, which has been applied in industry product, transport, electric power, office building and community and provides more and more help for people. The video surveillance system in the domain of education is studied as the actual exmple in this desertation, the key techniques of which mainly are solved including video compression, surprised object extraction and matching.The selection of education domain is decided by the requirement of usage background, and the education examination online monitoring system is taken as the actural studied object. The reason is that the behavior of fraudulent practice has become more and more concealable and the measures of them became technical. In order to efficiently beat the fraudulence of examination and improve the management efficiency, the examination center of ministry of education has put forward the education examination online monitoring system including the techniques of monitor, communication, process of audio and video, examination management would be constructed to cover with the examination institutions and spots.The video surveillance slowly becomes dependant on IP with the development of IP technique. The process of video is the kernel content, especially the selection of video coding standard would decide the performance of the whole system. The previous standards such as MPEG-2 and MPEG-4 don't meet the requirement of intending monitoring system that needs transfer magnitude of video through limited bandwidth. AVS and H.264 are the new generation video coding standards. Now AVS is being established and populazied, and H.264 possesses the absolute advantage in compression efficiecy, so some key techniques of new surveillance surround H.264 in the dissertion.H.264 has higher arithmetic complexity while it provides higher compression ratio. Its complexity is twice than the one of MPEG-2, which affects its real-time realization in a certain extent, and it needs higher process ability to the coding system. Motion estimation is the most complicated part in the H.264 encoder, which accepts multiple block modes, multiple reference frame and sub-pel interpolation etc. Its computation occupies most of the encoder. So the fast and highly efficient motion estimation is the most important algorithm to solve real-time coding.To this question, a motion estimation algorithm integrated with mode decision and multiple reference frame selection is proposed. This algorithm combines motion estimation with mode decision and multiple reference selection, and consider integer transform, quantification, vector predict, search range, reference frame and the correlation between block modes. It respectively process rational motion estimation method for each block mode and each reference frame. While maintaining coding efficiency, this algorithm can improve the coding speed greatly. Excpted that, the encoder is further optimized, which can realize the real-time coding of H.264 Baseline Profile encoder.When the video stream data arrive at the end server or analysis sever, the obtained image information need be analyzed at once, especially the moving object need be quickly exacted. At present, there are two methods: one is processed in pixel domain, which analyzes the decoded and reconstructed image by pixel unit. This method consumes more time, and usually need improve the performance of algorithm to meet the real requirement; the other is processed in compression domain, which can utilize some information of compressed bit stream to directly extract the moving object, for example the coding information or object coding information. These informations can directly be used to speed up the object exaction. This method has better practicability, but it is more complicated and still is being researched.To this question, firstly a real-time moving object extraction and matching algorithm in pixel domain is proposed. The motion vectors are cached while decoding, and used to fast label the moving connection region. After these works, the algorithm checking background is accepted to remove the background information, and the moving object is extracted. Furthermore an improved multi-resolution hausdorff distance algorithm is provided to match moving object. The experience results of this new extracting and matching algorithm show that it can quickly and efficiently extract and match many kind of moving object and meet the real-time requirement.Secondly two novel video object extraction algorithms based on MPEG and H.264 compression domain are proposed. The object extraction algorithm based on MPEG domain accepts a multi-resolution process method, and the experience results show that this method can achieve preferable effect. Excepted that, a global motion estimation algorithm in H.264 compression domain is provided, which can accurately estimate global motion, and the motion vector accumulation and projection algorithm is proposed based on above global motion estimation. The experience results show that all these can accurately compute the global motion and realize real-time object extraction.At the last, the streaming process of H.264 is introduced and analysized. And an actual prototype system of monitoring with multi-layer scheme is putted forward based on above research fruits, witch not only provides a platform for research, but also coaches the construction of the new online monitoring system.