Motion Analysis And Processing In Video Scenes

Analysis technology of moving objects in video processing has been utilized widely nowadays. In general, analysis of moving objects can be classified as detection, tracking, recognition and behavior understanding of moving objects. Detection and tracking of moving objects is the basis of others which pay great attention to such studying contents as extracting regions of interested moving objects from series of images, analyzing character of moving objects in order to match character and track, estimating parameters of movement and so on. Behavior recognition is analysis technology which makes use of computer language to define target behaviors according to different events and recognizes behaviors in specified scenes on the basis of detection and tracking of moving objects.This thesis builds a video surveillance system for recognition of tailgating behavior on the basis of summarizing some leading object detection and tracking algorithms. In a meanwhile we do a deep research in background modeling, shadow removal and target tracking related algorithms and improve some basic methods before presenting a definition of tailgating behavior for recognition. It is inevitable for some backgrounds to be influenced by weather especially in those outdoor scenes. Even if a robust background modeling method can not obtain dynamic background from rain or snow climate. This thesis applies algorithms of fast-moving object detection and removal as a basis to rain detection and removal in image background innovatively.The work this thesis done is as following:1. This thesis analyzes common background modeling methods at present and utilizes Gaussian Mixture Background Model for dynamic background in image sequence. Thanks to little color correlation between adjacent pixels in Gaussian Mixture Model, we present an improved method to update Gaussian parameter to make Gaussian Mixture Modeling more efficiently in real-time application.2. Because shadow point always misjudges as objective point in motion segmentation and object detection, it is crucial to detect moving shadow in dynamic scene analysis. In our anti-tailgating surveillance system commonly used shadow detection method based on normalized rgb color space combined with Gaussian Mixture Shadow Model to detect moving shadow.3. One of the main difficulties of the tracking process concerns the partial or total temporal occlusions of the objects. Therefore, we establish a tracking strategy to divide tracking process into seven different events to handle with. Objects can be divided and merged in tracking process which gives a high requirement for object matching. One character matching method based on color histogram similarity has been used to make objects matching in image sequence. In order to recognize tailgating behavior, definition of this behavior and detailed tailgating recognition process are given in the thesis.4. For the purpose of making surveillance system suitable for outdoor scenes, one kind of background moving objects, rain drops has been studied and we develop a rain detection and removal method from video images. Based on analysis of rain's physical and optical model in camera we detect rain spots taking advantage of improved frame difference method and K-means clustering method respectively and draw comparison. At last, we present a more effective rain removal method than classical.