Motion analysis of articulated objects for optical motion capture

Abstract

This thesis deals with the problem of controlling an articulated human figure using optical motion capture. Several systems have been developed, most of which track small reflective markers and the resulting 3-D trajectories are then applied to the human figure model. These systems copy human motion for the animated characters. The previous systems, however, require a very time-consuming manual post-processing to recover the trajectories when some of the markers are lost from the view. To alleviate the difficulty in the tracking process, we propose a model-based tracking algorithm for optical motion capture. The control problem of the model-based tracking is that of determining the 3-D position of a model segment along with its posture, i.e. the relative position of its parts from the frame by frame analysis of a sequence. The articulated object is modeled using the volumetric hierarchical model which includes several markers on the body segments. The analysis of a frame provides the information necessary to update the posture and position as well as the translational and the angular velocity of the body model. The validity of the updated model is verified by comparing its projection on the image planes with the position of the marker. We propose a hierarchical Kalman filter (HKF) to incrementally estimate the model parameters, and several error reduction techniques, which enable us to treat the problem in a single unified framework. Our formulation avoids the multiview correspondence and the model-feature matching problem as well as the inverse kinematics procedure. We demonstrate that our formulation works well through several experiments on both synthetic and real images. The results also verify that the model-based algorithm works well on the tracking of multiple rigid objects. Furthermore, we present a realtime animation technique by using the captured motion. It provides us a multi-level control mechanism of the motion resolution by changing the uncertain...