Shape Processing In Human Motion Analysis

Recovery of 3D body pose is a fundamental problem for human motion analysis in manyapplications. This paper introduces a new silhouette-based framework for inferring humanpose from sport video and pays more attention on how to retrieving 3D body pose bymatching 2D silhouette. There are three main results which are listed as following:1. The smoothness and noise reduction in silhouette based on Level Set.2. The shape matching based on EMD and LSH methods.3. The silhouette modeling based on Shape Context.By many experiments, it is demonstrated that once the silhouettes abstracted from videoframe is accurate enough, the accuracy of reconstruction results can be 97%. To accelerate thematching process, the paper uses PCA (Principal components analysis) to reduce thecomputation of complexity.