Motion capture from single video sequence

Abstract

3D human pose reconstruction is a popular research area since it can be used in various applications. Currently most of the methods work for constrained environments, where multi camera views are available and camera calibration is known, or a single camera view is available, which requires intensive user effort. However most of the currently available data do not satisfy these constraints, thus they cannot be processed by these algorithms. In this thesis a framework is proposed to reconstruct 3D pose of a human for animation from a sequence of single view video frames. The framework for pose construction starts with background estimation. Once the image background is estimated, the body silhouette is extracted by using image subtraction for each frame. Then the body silhouettes are automatically labeled by using a model-based approach. Finally, the 3D pose is constructed from the labeled human silhouette by assuming orthographic projection. The proposed approach does not require camera calibration. The proposed framework assumes that the input video has a static background and it has no significant perspective effects and the performer is in upright position.