Research On Moving Object Extraction On Videos And Its Application On Security

Moving object detection in video sequences has been a widely concerned research subject. Extracting moving object accurately has significance to later analysis and processing. The thesis mainly focus on the problem of moving object extraction on security door surveillance video, we study an improved background method based on texture information and an improved frame-difference method based on edge information respectively.Moving object extraction in video surveillance is an application subject with specific problems, the main problem is the interference caused by variation of the ambient light, including slow changes, abrupt changes and uneven changes. The main work of this thesis is to overcome the variation of the ambient light and accurately extract the single moving object from the security door surveillance video. After the discussion about several popular moving object extraction algorithm such as background difference method, optical flow method, frame-difference method and statistical method, this thesis focused on studying an extraction algorithm to deal with the interference caused by ambient light varience in the security door surveillance video.The thesis firstly studied an improved background difference method based on block varience; The method considers the local information,using normalized varience as feature to build background model, it can adapt to the variance of ambient light in some degree; The experimental results proved the feasibility of this method. But the improved background method still has some problems, the extraction results are not accurate and the ability to deal with ambient light variance is limited. To improve extraction results, the thesis then studied an improved frame-difference method based on edge information, the method is fast and insensitive to the variance of ambient light. The experimental results prove that the improved frame-difference method can be used as a follow-up process after the block variance background method to improve the extraction results.