Research Of Moving Object Segmentation And Tracking Based On H.264/AVC Compressed Domain

Moving target detection and tracking from a video stream, means detect moving objects in video sequences, and track the moving target along the timeline. This technology is widely used in surveillance, video retrieval, pattern recognition and other related fields. It also been concerned by academic, because of its closely related with modern life.The research on moving object has been concentrated in the pixel domain. The high computation time cannot meet real-time requirements of urban life. With advances in multimedia and video coding standard, more and more video information in a compressed format to store and transport, so it is significant to research the technique of moving object in compressed domain that is especially useful in real-time environment, because there is no need to fully decode compressed streams and the data need to be addressed greatly reduced.The research on moving object detection and tracking n compressed domain is followed the footsteps of video coding standards. Compared with previous coding standards, H.264/AVC is absolutely concerned by people, but the most method is as the pixel domain. This paper based on H.264/AVC coding standard; study the codec, analysis the generation of the moving information theory. From the respect of moving object detection and tracking to re-granted the motion vector field and the macroblock partition type to a new meaning. Then proposed a new method of moving object detection and tracking in H.264/AVC compressed domain. This paper has three parts:1. Moving object detection, use the motion vector and macroblock partition type as motion information, an algorithm based on adaptive threshold is proposed. From two aspects to measure the state of moving object, that are movement features and texture features. Then through horizontal and vertical projection method to determine the location and number of moving objects. Experimental results for several video sequences with different characters, and subjective and objective dates be used to analysis the results.2. Moving object segmentation, use the motion vector and macroblock partition type as motion information, presents a novel approach for moving object segmentation, which based on ant colony clustering algorithm. The algorithm takes advantage of the characteristics of motion vector field. Use of inherent characteristics of motion vector field to selects cluster centers automatically, while use macroblock partitioning type as supporting information to correct the result. Experimental results for several video sequences with different characters, and analysis experiment results with the computation time, accuracy of segmentation and so on.3. Moving object tracking, use the motion vector as motion information, presets an approach for moving object tracking, which based on mean shift algorithm. The algorithm establish target model with direction angle of the motion vector. When the target and the background have similar color distribution form, the target is not lost. Experimental results for several video sequences with different characters demonstrate that the proposed approach can track moving object effectively.After all, only need to extract motion vectors and macroblock partition type from the stream in the whole process. Without decode the entire stream. In the case of reduce dates obviously, a real-time requirement is achieved.