Constraint-conscious smoothing framework for the recovery of 3D articulated motion from image sequences

3D articulated motion is recovered from image sequences by relying on a recursive smoothing framework. In conventional recursive filtering frameworks, the filter may misestimate the state due to degenerated observation. To cope with this problem, we take into account knowledge about the limitations of the state-space. Our novel estimation framework relies on the combination of a smoothing algorithm with a "constraint-conscious" enhanced Kalman filter. The technique is shown to be effective for the recovery of experimental 3D articulated motions, making it a good candidate for marker-less motion capture applications.