In this paper, we investigate articulated human\nmotion tracking from video sequences using Bayesian\napproach.We derive a generic particle-based filtering procedure\nwith a low-dimensional manifold. The manifold can be\ntreated as a regularizer that enforces a distribution over poses\nduring tracking process to be concentrated around the low dimensional\nembedding.We refer to our method as manifold\nregularized particle filter.We present a particular implementation\nof our method based on back-constrained gaussian\nprocess latent variable model and gaussian diffusion. The\nproposed approach is evaluated using the real-life benchmark\ndataset HumanEva. We show empirically that the presented\nsampling scheme outperforms sampling-importance resampling\nand annealed particle filter procedures.
Loading....