Estimating and Recognizing 3D Articulated Motion via Uncalibrated Cameras

The main goal of this effort is to develop new algorithms for 3D articulated structure and motion estimation, from one or more uncalibrated video streams. Articulated motion is exhibited by jointed structures like the human body and hands, as well as linkages more generally. In this project, 3D articulated structure and motion estimation algorithms will be developed that can automatically initialize themselves, estimate multiple plausible interpretations along with their likelihood, and provide reliable performance over extended sequences. To achieve these objectives, concepts from machine learning, graphical models, multiple view geometry, and structure from motion will be employed. The proposed research effort will focus in two main areas: (1) 3D articulated pose estimation given video obtained from uncalibrated cameras, (2) statistical learning models that capture the dynamics of articulated motion, to provide top-down guidance that is needed to improve the pose estimation and to allow motion recognition. Effort will also be devoted to investigating improved features and image segmentation methods for use in the front-end system. The developed methods will be tested on videos depicting motion of the human body and the human hand, where ground truth is available for quantitative comparison.