Human Body Recognition And Tracking Based On Sequent Images In Complex Background

The analysis of sequent images becomes more and more valuable. It has been put attention to by many subjects such as Computer Vision, Artificial Intelligence and Pattern Recognition est. It refers to that distill the features about the objects include object detection, tracking and recognition of the object and so on from the static sequent images. It can be used widely in such as virtual reality, automatic inspection and interaction computer.Usually, a full process of movement analysis covers the following stages: object detection, object recognition, object tracking and understanding & analysis of the movement. Not all of the stages must exist at the same time. However, object detection and tracking is necessary commonly. Thus emphasis is set on the two stages, at the same time , introduction of recognition is of presentation here.Separating the moving object from the sequent images calls object detection. And the process is important to other courses. Because of complexity in practice such as changing of light, shadow existing and disturbed from other things, it is difficult in the process. Now ,the arithmetic of moving detection is almost based on kinds of hypothesis, and it is uneasy to find an algorithm to apply all kinds of circumstance. There are three types algorithms to detect object from the background: background image difference, difference in Image Sequences and optical flow. Here, method of background image difference is adopted, and the background is recovered by blocks of sub background images, and is updated timely.Before recognize the object, following process is necessary: abandon the objects that were detected by mistake, filter the noisy dots and fill in the little holes. In this paper, the object is classified into human and inhuman. The principia is that if the object is totally in sight, the proportion of human body's height and width is used to justify whether it is a human. Otherwise human body's model is used to match the object. When the result is beyond a threshold, the object is marked as a human. If the object is not in sight fully, when a face is detected, it must be a human, or else, partly human body's models are used to mach the object.When a human body is detected, the object is tracked in the next process. Here the method of CamShift is presented. It uses the information of color to track the object, timely and robustly. But when the object moves quickly and stops for a moment, the object will be lost. In this paper, CamShift and Object detection are used together to track the object. And more than one object can be tracked.Analysis shows that the algorithms in this paper are effective, object can be detected and tracked robustly.