Human Body Detection And Tracking Based On Head And Shoulder Feature

The detection and tracking of human motion are key to the visual analysis of human movement, It involves wide application foreground and great economic value in smart security surveillance, advanced user interface and the details analysis of human movement, etc. The detection of head and shoulder is not easily disturbed by other features (such as the gait, the clothes color, environment) and it is proved advantageously for the precise detection and tracking for moving bodies. The thesis is about the research on detection and tracking of moving human body in the video sequence captured by single fixed camera. The main work is as below.On extracting of moving object, the thesis summarizes the current usual method on detecting moving object, and adopts the background subtraction to extract foreground moving object, this method could extract the moving object's whole feature precisely, the calculation is moderately complicated, the running time is short. After the background subtraction, it makes shadow removal, get the precise object shape, and eliminate the shadow. The image of eliminated the shadow is processed by mathematics morphologic operation and the connected-component identifier. It provided advantage for precise detection and human body object tracking in the future.On human body moving object detection, the top of the moving object is located by vertical projection histogram, and the head and shoulder is located according to top position. The binary image sample database is built based on the whole body, the head and shoulder when the human body is moving to 6 directions, and the head and shoulder region's features are extracted by Principal Component Analysis (PCA). If it is greater than the defined threshold of the region is classified to head and shoulder region, and it is not the moving human body.On the aspect of tracking moving human objects, the motion state of moving human object is estimated by Kalman filter in sequential images, the input is the central position of human head and shoulder and the trajectory is obtained by predicting the position of moving human object in the next frame. This method can reduces the noise disturbance and the searching time of the feature extraction, the experimental results show that it is efficient and effective.