Dynamic Bayesian Network Model Based Golf Swing 3D Reconstruction Using Simple Depth Imaging Device

The simple depth imaging device gains more and more attention because of its lower cost and easy- to-use property compared with traditional motion capture systems. However, this kind of devices lack the basic data condition of 3D motion reconstruction due to low resolution, occlusions, and mixing up of body parts. In this paper, a Dynamic Bayesian Network (DBN) model is proposed to describe the spatial and temporal characteristics of human body joints. The model is based on fusion of the parent-child characteristics of joints and multi-order Markov property of joint during motion. A golf swing capture and reconstruction system DBN-Motion (DBN-based Motion reconstruction system), is presented based on the DBN model and the similarity of swing with a simple depth imaging device, Kinect, as capturing device. The proposed system effectively solves the problem of occlusions and mixing up of body parts, and successfully captures and reconstructs golf swing in 3D space. Experimental results prove that the proposed system can achieve comparable reconstruction accuracy to the commercial optical motion caption system.