Key Pionts Of Human Body Location Based On Single Depth Map

Recently, with the development of image processing and artificial intelligence, thetechnology of human motion analysis and understanding has made great strides, and it hasbeen used in many fields. The premise of human motion analysis and understanding is tocapture the state of human’s action correctly. There are many problems in the traditionalmethod for the capture, such as can’t automatical initialize, can’t recover from the trackfailure, can’t solve the occlusion problem, or there are many constraints on the participants,and so on.If the key points of human body can be determined from a single image, the problem ofinitialization and self-recovery can be solved. As eyes can get enough information fromsingle image to realize occlusions, it has reason that computers can make it, too.There are two phases to obtain key points of human body from the image: the divisionof the body and background, and the body key point location. Kinect, a depth camera fromMicrosoft, which can catch the distance of objects to the camera, provides a perfect way tosegment the body from background. Meanwhile, the depth map can also reflect the humanbody’s motion when occlusions occur. The location of the arm can be identified by thesudden change of the depth value in the body region when it is in front of the body. To usedepth maps can also get rid of the influence of people’s clothing and light conditions.In addition, the skeleton can represent the body’s structure and movement visually.Making use of mathematical morphology approaches to reduce the image’s boundary andinternal noise, it can get the skeleton of body more accurately. And Kinect can providedepth information to solve the problem of getting the skeleton when there’re partialocclusions. After locating the point of maximum curvature in the skeleton line, which isthe key point, the human body state parameters can be absolutely obtained.To locate the human body’s key points from a single depth map, it is better to used inmotion tracking to solve the problem of automatic initialization and self-recovery.