Technique Of Motion Detection And Objects Decision In Visual Surveillance

With the development of electronic technology, communications and computer vision, the researches and applications of the intelligent visual surveillance get more and more attention of people. As the base of video processing in surveillance system, motion detection and objects decision is one of the most important technologies.This dissertation is making analysis of the algorithms involved in motion detection. A method of motion detection based on temporal differencing and background differencing is proposed. The method making detection with the background difference method based on mixed Gauss model at first, then judge the result by cumulate three-frame temporal differencing. This moving object detection method can avoid error detection and illusory detection that often happened in video surveillance system used presently, and can detect in time with global-illumination changes.Objects decision can comprehend in classify the moving objects into semantically meaningful categories, and associating the correct object class label with the region of homologous. On the basis of analyzing image features in moving target, a method of objects decision by shape features of object is proposed. The method defined shape features for moving object at first, then acquire the parameter for classify with feature extraction. A multi-class classifier is constructed based on support vector machines, and the classifier is trained with small sample. Simulation results show that this classifier can class person, vehicle, dog and person with autocycle preferably. The correct test classification rate attains 97.38%...