Memory-based Particle Filtering

Summary:

Memory-based particle filtering (M-PF) has been developed to achieve robustness against complex motion and recoverability from tracking failure in visual tracking. To that end, we eliminate the Markov assumption from the previous particle filtering framework and predict the prior distribution of the target state from the long-term dynamics. More concretely, M-PF stores the past history of the estimated target states, and employs random sampling from the history to the generate prior distribution; it represents a novel PF formulation. Our method can handle nonlinear, time-variant, and non-Markov dynamics, which is not possible within existing PF frameworks. Accurate prior prediction based on proper dynamics model is especially effective for recovering track loss, because it can provide possible target states, which can drastically change since the track was lost. We target the face pose of seated humans in this paper.

Some demo movies are shown below. Please refer the proceedings paper and panel for details.

This video shows how the proposed tracker behaves, including initialization, data accumulation, and prior predictions. White mesh indicates the tracked face pose. Green points and red points denote sampled particles.

Publication: