A simple and robust approach to drift reduction in motion estimation of human body

Abstract

Incremental tracking of the human body is performed by accumulating pose displacements frame by frame over the initial pose at the starting frame. The initial pose is measured by fitting an articulated model to the human body in an image, and the pose displacement can be estimated from the intensity difference between successive frames. However, the incremental approach faces a difficulty, namely, drift by accumulation of estimation errors of pose displacements. This paper proposes a simple and robust method of drift reduction. The proposed method measures a pose by model fitting even to the final frame, and calculates the drift between the final pose and the pose obtained as a result of accumulation. The mean drift between the frames is then calculated from the drift at the final frame, and the pose is corrected by subtracting the mean drift from the pose by accumulation in turn. Several iterations of motion estimation and pose correction make the model fit the human body in the intermediate frames. Pose correction can be performed simply by solving a system of linear equations in three unknowns. Several experiments in body tracking prove the proposed method to be effective. © 2006 Wiley Periodicals, Inc. Electron Comm Jpn Pt 3, 89(8): 39–52, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjc.20264