Video-Based Human Motion Estimation System

This paper presents the system designed to estimate body silhouette representation from sequences of images. The accuracy of human motion estimation can be improved by increasing the complexity of any of the three fundamental building blocks: the measured data, the prior model, or the optimization method. The vast majority of existing literature on human motion estimation has focused on just one of these building blocks: improving the methods for optimization, also called inference. In contrast, our approach seeks to explore the hypothesis that the other two building blocks are critical components, using extremely high accuracy measured data and shape of body motion priors, so that the objective function is more precise and less noisy, resulting in an easier solution. Our main goal is to develop a new module for extracting accuracy measured data from video imagery