Dimensionality Reduction for Articulated Body Tracking

We present a novel combined approach for 3D body part tracking using multiple cameras, called GPAPF. This approach combines annealed particle filter body part tracker with Gaussian process dynamical model (GPDM). We use GPDM in order to reduce the dimensionality of the state vector. This reduction improves the tracker's performance and increases its stability and ability to recover from loosing the target. We also present a way to create a latent space, which is rotation and translation invariant. We compare between GPAPF tracker with an annealed particle filter and show that our tracker has a better performance even for low frame rate sequences.