| With the development of telecommunication, computer, television broadcasting and micro-electronic technique, the digital video processing and communications have gradually become the common research fields of academic and industrial groups. Especially the international standards about the digital video transmission: H.261, 263 and the international standards about multimedia compression: MPEG-1, 2, 4, 7 are completed. The high-speed processing chip and mass storage appear. The digital video products become more popular and important in social life, production, business, scientific research, education, military, and so on. This technique takes pivotal effect in the applications such as high definition television, videophone, videoconference, machine vision, surveillance, video guiding. In this dissertation the motion analysis, object segmentation, compression and transmission technique of the digital video are the main topics. The theory system, mathematic model, technique principle, realization scheme, algorithm and applications are discussed thoroughly and carefully. After the recent results of research at home and abroad are summarized, some new theories and techniques are put forward creatively in this dissertation. They will benefit the development of the digital video technique.The video segmentation and object extraction are one of the premises for the efficient representation, compression and transmission of the digital video. They require the motion analysis for the video. So in the first, the basic principle for motion analysis model is summarily discussed, including the motion model of 3-dimensional objects, the projection model of cameras and typical analysis technique for motion fields. The optical flow estimation technique is one of the important motion analysis techniques, because of its wide use. But Sometimes using the optical flow equation directly for high-speed motion analysis in successive frames of the digital video, we will find the obvious error in the final results. The combination optical flow estimation with displacement pre-compensation is put forward to eliminate this kind of error for precise results. At the same time, the factors, which introduce the error of temporal differential estimation, have been analyzed. The successively approaching calculation model for the optical flow estimation is developed to reduce this kind of error. It greatly improves the accuracy of the optical flow estimation. The theoretic and experimental results have proved that the above two techniques are better than usual obviously.On the basis of video motion analysis, the video segmentation and object extraction are discussed from two aspects: One is the background subtracting method. The other is based on the gray morphology.For the background subtracting method, the background image is required. So the video decomposing and description model about the background and foreground is brought forward. Then the background image is estimated. The compensation for the plane and non-plane background is analyzed in the condition of the moving camera. The premise of plane background in the global motion estimation and compensation is proved from the theoretical and instance analysis. The coding scheme based on the layered background is put forward so as to realize the high efficient representation and compression. It also benefits the moving object detection and segmentation.After the background image is obtained, the video segmentation and object extraction can be completed conveniently. And the moving objects stream analysis is discussed. The similarity operator for spatio-temporal location and gray characteristic is utilized to track the moving objects. On the far watching, the moving objects stream is decomposed according to the orbits. On the near watching, the sample picture for the moving object is grasped at the set location. In the end the traffic video sequence on the road is used in the experiments and the result is satisfactory.In the gray morphology method, the spatio-temporal potential e...
|
PDF Full Text Request