Prediction of Body Configurations and Appearance for Model-Based Estimation of Articulated Human Motions

In this paper we address the problem of predicting human body configurations and appearance for the task of humanmotion tracking in video. Unlike many previous approaches that relied on simplified body models, we are able to use detailed models of body shape. Having rendered the body model onto the current video frame, we make use of the optical flow information to propagate the body configuration to the next image frame. Parameters are estimated via a robust optimization technique which is based on a Kalman filter approach to the prediction of body configurations in the space of articulated body motions. Results are shown demonstrating the effectiveness of our approach.