Research On Vision-based Human Motion Tracking And Recognition

Vision-based human motion analysis plays a very vital role in intelligent video surveillance system. Because of the application of vision-based human motion analysis technology, moving human can be detected and tracked in cluttered image sequences by video surveillance system, and it alerts to protect the safety of persons and property when abnormal conditions are found. Therefore, in order to achieve intelligent video surveillance of laboratory, it is essential to make intensive studies in the theory and technique of vision-based human motion.By analyzing and comparing the merits and demerits of the commonly used moving object detection, tracking and target recognition methods, this thesis presents moving human detection algorithm, tracking and recognition algorithm in complex and cluttered scene. In the part of detection, foreground region is extracted by background subtraction algorithm based on Gaussian mixture model. Also, the shadow detection and elimination is implemented in the HSV color space, and some after treatment operations are performed. This thesis employs a partial Hausdorff distance tracking algorithm which combines with Kalman filter for no occlusion among moving humans. For the occlusion issue, a tracking and recognition algorithm is designed by parting occlusion process and defining the conditions of occlusion generation and end. The view, tracking the whole moving target, is used after moving humans overlapping and shaping a block. The original method of establishing color histogram is improved, thus a method of computing the number of pixels based on the weight is adopted to establish color histogram, and Bhattacharyya distance is computed to recognize moving humans. Finally, according to laboratory environment and the requirements of software design, this thesis designs a supervisory control software for identifying suspicious person on the basis of the proposed approaches.This thesis conducts experiments on moving human detection, moving human tracking and recognition. The results show that the proposed algorithm can accurately dctect and track moving human in a complex scene. During testing the supervisory control software for identifying suspicious person, the results indicate that when someone enters illegally, that person is marked in the image and the software gives an alert.