Research on Real-time Estimation for Human Pose

This study addresses a method for real-time estimation for human pose. Firstly, depending upon his feature, image segmentation is performed on a human dynamic image preprocessed. His body composition is secondly sampled and its key points are correspondingly selected. His human skeleton and contours are separately sketched according to such key points' positions and his feature extracted. Finally, distance matrices can be built from the distances between the key points of human skeleton in order to realize image recognition with the result that such pedestrian pose is capable of being estimated successfully. The experimental results shows that his method has a better real-time capability and usual poses can be estimated to some extent.