The Research Of Video Moving Objects Extracting Method Based On DWT

Moving object extraction in the video sequence is one of the main content in the field in computer vision research, as well as digital image processing and pattern recognition. only based on the extracting of the video moving target that the research works such as objects classification, tracking, and content analysis could be carried out. Therefore, the video moving object extraction technology has great significance both in theoretical research and practical application, is also a key technology in the field of computer vision and video processing. This dissertation presents some efforts on extracting moving objects from monocular video captured by static camera, and the main contributions of my work include:Due to the disadvantage of large volumes data of video image sequence and information redundancy between images, based on the hypothesis that the background pixels intensity always appear in image sequence with maximum probability,a novel moving objects segmentation method combing DWT and background reconstruction was presented. This approach extracted the approximate components of video images sequence firstly using the DWT(discrete wavelet transform),which resulted in the image size shaped into1/4of original image to improve the processing efficiency; then reconstructing the background according to the pixels clustering combing dual-threshold with similar classes merged method; at last, extracting the moving object with the application of the background subtraction.Building on the above raised method, after analysis the key parameters in the clustering algorithm--threshold, the shortcomings of fixed threshold in the clustering was pointed out. An adaptive online clustering moving object extraction algorithm was developed to address this problem through the intensive study. In view of the problem of imperfect effect evaluation system of reconstructed background,this paper verified background accuracy by quoting the objective evaluation criteria of image matches.Experimental results indicated that the proposed algorithms could extract moving objects accurately and rapidly with good robustness, even in the cases of environments change, moving targets roundabout, multi-targets moving.