Moving Human Detection And Tracking Research Based On Video Frames

Moving target recognition and moving target tracking based on image sequence is one of major research directions in computer vision. It is the basic technology in the application of intelligent surveillance, moving robot navigation, precisely guided weapon and human-machine interface. It includes three parts:(1) moving target detection; (2) human recognition; (3) moving human target tracking.In this text, a human detection method in video sequence based on background subtraction and closed contour fitting has been proposed if the background image has been acquired. Firstly, after having the background image, seek for the maximal connected domain of the original foreground objects; Secondly, extract contour of the original foreground object through weighted gauss filters; Finally, fit the closed contour of the original objects and extract them, a method of 2bit area chosen has been proposed in order to seek for end points of one segment of a curve, then according to the criterion of the minimum distance between class structure closed contour, extract final foreground objects by using frame difference algorithm, and label the locations of the objects.In order to recognize human target, a method based on deformable ellipse template and cascade matching is presented according to the characters of human’s head and shoulder. Construct ellipse templates of head and shoulder, and express by parameters. Change the parameters of template according to the real target in the image. Then recognize human target by computing the Euclidean distance of the gradient histogram between the template and the real target.A dynamic probability classification model is proposed for moving human target tracking. Compute the classification probability matrix of the model according to the maximum posterior probability. Then generate a set of state sequence, and the sequence contains moving direction message of the human target.A single-target/multi-targets recognition and tracking system based on image sequence is realized according to the research. The system is developed on Matlab 7.11, and it contains load-video-sequence module, target detection module, target recognition module and target tracking module. Experiment results shows that the system promoted efficiently and robustness in recognition and tracking process.