Real-time gesture recognition system based on Camshift algorithm and Haar-like feature
This paper designs and implements a following robot system based on practical monocular vision sensor. The system can autonomously follow a specific target human body as needed. In the initialization phase, the system uses the background difference method to obtain the motion region and extract the target. In the second phase, continuously adaptive mean-shift algorithm (Camshift) is used to the target tracking, and Kalman filter is used as the starting point of the Camshift to compensate the target shift caused by the camera movement. In the third phase, the feature points of the target human body are extracted and the pose parameters of the target human body relative to the robot location are obtained by the 3D pose estimation algorithm (pose from orthography and scaling with iterations, POSIT). Experiments show that the system has a good stability, and can effectively implement the target human following.