CS 2523 OMR Final Project Report : Human Body Pose Estimation

H UMAN Body Pose Estimation (HBPE) is the process of identifying how a human body and/or its individual limbs are configured in a given scene [9]. The importance of such task has been identified by the computer vision community which has dedicated a large amount of effort towards finding an effective solution. A wide variety of techniques have been proposed in an attempt of finding the one which performs better in a wide variety of scenarios. However, the ultimate solution is yet to come and research in this area is quite active these days. It is not surprising to find such a great interest and dedication towards the solution of this problem, not only because of the extensive number of applications (medical diagnosis and treatment, sports training, surveillance, study of human behavior, motion capture, among many others), but also because it presents huge difficulty, as people can describe a large (and perhaps unpredictable) set of poses, on top of the natural variability that comes from humans themselves. The initial goal of this project was to build a system to (partially) solve the problem of HBPE based on [19]. However, it rapidly became clear that neither the simple implementation of a system would allow me to explore the richness of the existing solutions in this area, nor I would be able to come up with mature ideas to incorporate to the implementation without a comprehensive literature review. At the end, the this review became a significant portion of the work of this project and this report will summarize important concepts I learnt from both the implementation and the literature review. The remaining of this report is organized as follows: A summary of the literature review is presented in Section 2, followed by the proposed methodology on Section 3. Section 4 shows some experimental results and finally Section 5 presents some conclusions.