3-D–2-D spatiotemporal registration for sports motion analysis

Abstract

Computer systems are increasingly being used for sports training. Existing sports training systems either require expensive 3-D motion capture systems or do not provide intelligent analysis of user’s sports motion. This paper presents a framework for affordable and intelligent sports training systems for general users. The user is assumed to perform the same type of sport motion as an expert, and therefore the performer’s motion is more or less similar to the expert’s reference motion. The performer’s motion is recorded by a single stationary camera, and the expert’s 3-D reference motion is captured only once by a commercial motion capture system. Under such assumptions, sports motion analysis is formulated as a 3-D–2-D spatiotemporal motion registration problem. A novel algorithm is developed to perform spatiotemporal registration of the expert’s 3-D reference motion and a performer’s 2-D input video, thereby computing the deviation of the performer’s motion from the expert’s motion. The algorithm can effectively handle ambiguous situations in a single video such as depth ambiguity of body parts and partial occlusion. Test results on Taichi and golf swing motion show that, despite using only single video, the algorithm can compute 3-D posture errors that reflect the performer’s actual motion error.