Abstract
The thesis comprises two part of research: human posture reconstruction and human motion recognition. In the part of human posture reconstruction, we propose a novel model-based approach to reconstruct 3D human posture from a single image. The approach is guided by a posture library and a set of constraints. Given a 2D image and the users label body segments of human figure and estimate root orientation in the image, a 3D pivotal posture whose projection is similar to the 2D human figure will first retrieved from posture library. To facilitate the retrieval process, a table-lookup technique is proposed to build an index structure of posture library. Next, constraints including physical and environmental constraints are automatically applied to reconstruct the 3D posture. Experimental results show the effectiveness of the proposed approach. In the part of human motion recognition, we use motion capture data to generate simulated 2D motion trajectory. Next, four recognition algorithms including Dynamic Time Warping, Support Vector Machine, Hidden Markov Model, and Dynamic Bayesian Network are exploited to recognize different motion. We will discuss the experimental results in depth and compare different recognition algorithms.