Studies Point To Rectify The Problem Of Human Bone Skeletal Depth Image-based Tracking

Human skeleton tracking is an important technique in computer vision and it is widely used in the fields of game, film and television, and so on.In this paper, the traditional model of skeletal tracking used multi-camera is changed and the skeletal tracking is obtained by using depth sensor. The skeletal tracking technique can capture in real time coordinates of the body’s20skeletal points by depth sensor. Its advantage is more convenient to get the depth images and skeletal point location information. But as some body parts are occluded, the skeletal points will appear inaccurate deviation or drift phenomenon. But as some body parts are occluded, the skeletal points will appear inaccurate deviation or drift phenomenon.Decision tree is a fast and effective classification methods as well as pruning when necessary, in order to improve the efficiency of the classification. The main properties of the selected tree is characterized by the depth gradient and location coordinates of the image. Random Forest classifier gradient features are extraced in different directions, and is trained from the perspective of pixels to achieve partial division based on Microsoft’s Kinect sample library.Aiming to the problems of the skeleton points drift when tracking human motion skeleton, a novel method of limited the part circle that base on Gaussian process regression correction (GRP)and Support vector regression (SVR) is proposed to redress the skeletal point coordinates. In this method, the20joints in Kinect SDK coordinates are rebuilt by filtering, human pixel classification and location identification. The method proposed can solve the problem of the body’s part occluded, and achieve excellent performance. The experiments show that the method is more accurate and reasonable.