Face Tracking Using Rao-Blackwellized Particle Filter and Pose-Dependent Probabilistic PCA
Paper in proceedings, 2008
This paper deals with face blob tracking, where face undergoes various pose changes. We propose a novel tracking
method to deal with face pose changes during tracking. In the method, tracking is formulated as an approximate solution to the MAP estimate of state vector, consisting of a linear and a nonlinear part. Multi-pose face appearance is modeled by locally linear models, and estimated by the probabilistic PCA for individual pose combined with a Markov model for pose changes. Shape and locations of face blobs and pose index are assumed to be nonlinear and estimated by Rao-Blackwellized particle filters (RBPF), which also enables separate estimation of linear state vector through marginalizing the joint probability. The proposed method has been tested for videos containing frequent face pose changes and large illumination variations, under 5 pose models (left, frontal, right, up, down), and the tracking results are shown to be robust to varying speed pose changes and with relatively tight boxes.
object pose model
object appearance model
Rao-Blackwellized particle filters
Markov pose model