SYSTEM FOR RECOGNIZING POSTURES OF MULTIPLE PEOPLE EMPLOYING OPTICAL FLOW DETECTION AND BODY PART MODEL

摘要

A body posture recognition algorithm comprises the following steps: acquiring an optical flow vector for an image in a video to obtain a moving object region in the image; and using a body part-based method to perform body posture recognition on the moving object region. In the above algorithm, optical vectors for a sequence of images are acquired; data and space constraints are used to detect a change in the optical flow vectors between the sequence of images; robustness estimation is used to optimize a parameter; iteration and rectification are performed to obtain optical flow information; sliding window detection is performed on an extracted moving object region; comparison is performed to obtain a degree of similarity between the image and a body part shape template; a tree structure is used to model a human body; comprehensive calculation is performed, by means of information transmission, on correspondence of the image to respective body parts, thereby locating a human body in the image; back transmission is used to locate body parts, thereby realizing estimation of a body posture; a correspondence relationship between moving regions is calculated, thereby recognizing postures of multiple people, calculating the number of people, and tracking people.