Awareness of 3-D Pose Trajectory in Video Contents With Optimal Control Refinement

Abstract

This study determines 3-D pose trajectories for video contents by developing a 3-D pose-tracking algorithm that is refined by an optimal control algorithm. The proposed method reduces accumulated tracking errors better and is more computationally efficient than local bundle adjustment approaches. Note that the proposed method is the first complete optimal control formulation of 3-D pose-tracking. Different from contemporary control-based approaches, this formulation does not use any simplifications or assumptions. As a result, the proposed method is not limited to 2-D, specific motions, or specific objects' geometries.