Real-Time Moving Target Detection Based On Multi-View Image Sequence

Moving target detection based on video sequences is a hot research topic in the field of Computer Vision, Video Surveillance, etc, and aims to extract the interested moving target from a video sequence using certain methods. As the basis process of the entire image sequence analysis, the test result of moving target detection will have an important impact on the subsequent processing work. Therefore, practical applications raise higher requirements on the real-time problem of moving target detection. For the real-time problem existing in current moving target detection algorithms, especially for the problem of moving target detection based on multi-view, after analyzing the existing variety of moving target detection methods and methods for efficiency optimization, this paper proposed a paralleled moving target detection method based on image block. It can improve the efficiency of moving target detection combining with the visible moving range bound of targets from multi-camera, and the spatial and temporal coherence of moving targets.1. Firstly, we do reseach on the existing methods of moving target detection based on video sequences, and find a suitable method for our target after analying the detection processes, advantages and disadvantages of these methods. In terms of the research status of real-time moving target detection algorithm, we detect the moving target based on image block, and use GPU and CUDA technology to achieve parallel processing.2. For multiple cameras scenes of moving target detection, the actual visible range of movement for the target is limited by locations of multi-camera. Therefore, reference to the construction method of Visual Hull, we do the intersection operation for view space of multicamera to get the max visible range of movement. Then we calculate the effective imaging area on each camera projection plane. For redundant blocks in non-effective imaging area, we avoid handling with these blocks, thus the detection process can be accelerated.3. The paper studies the characteristics of interframe coherence of moving targets and the spatial and temporal coherence of result image data. Then based on image block, we use the detection result of the previous frame to estimate the detection range of the current frame. By organizing appropriate data structure, we can detect moving targets from successive frames quickly.Finally, We realize the algorithm put forward in this paper on the experimental platform we provide. And we do the comparison of experimental data and performance analysis between our method and traditional moving target detection method based on pixel. The experimental results show that, the proposed method can better improve the detection efficiency, and enhance the real-time nature of detection, in the premise of ensuring the detection accuracy.