|
Physical Simulation for Probabilistic Motion Tracking
|
We propose a full-body 3D physical simulation-based motion prior for human kinematic tracking that explicitly incorporates motion control and dynamics procedures into the Bayesian filtering framework. Most prior approaches to tracking have concentrated on efficient inference algorithms and prior motion models; however, few can explicitly account for physical plausibility of recovered motion. Towards greater physical plausibility, we consider a human's motion to be generated by a ``feedback control loop''. In this control loop, Newtonian physics approximates the rigid-body motion dynamics of the human and the environment through the application and integration of forces. Simulation-based prediction allows for faithful modeling of human-environment interactions, such as ground contacts, resulting from collisions and the human's motor control. However, such prediction introduces additional latent state variables related to the human's control policy and motion control. For efficient inference, we introduce an exemplar-based motion control strategy using motion capture data of known actions. An implementation of our approach demonstrates the ability to recover the physically-plausible kinematic and dynamic state of the body from monocular and multi-view imagery. We show, both quantitatively and qualitatively, that our approach performs favorably with respect to standard Bayesian filtering methods.
Video
Sponsors