Human motion estimation from a single view
Yi Sun, AN Hong-jie, Chengyi Cui · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2000
Human motion analysis is receiving increasing attention from biomedical image processing researchers. In order to reflect the human motion in reality, the body's structure is recovered using its 2D model. This paper proposes a practical system which tracks human motion automatically. The major processing units are as follows: 1) coarse matching between real image and 2D model, 2) fine matching between real image and 2D model, 3) the formation of body structure from 2D model sequence. We first segment a human body from stationary background. Then prior posture database is established, and the primary posture in image sequence can be estimated coarsely by comparing posture in database and real image series. After this, in the precise adjustment stage, precise matching can be obtained by the criterion of region overlay between image sequence and 2D model. Finally, structure of human body is recovered by adjusting parameters of 2D model series, in the above process, the sizes of the body parts are measured manually from one of the picture of real images. Finally, 2D model can be established, and skeleton or frame representation of human body movement is given. In the end of this paper, future directions are suggested for further improvement.