People Tracking with the Laplacian Eigenmaps Latent Variable Model

Reliably recovering 3D human pose from monocular video requires models that bias the estimates towards typical human poses and motions. We construct priors forpeople tracking using the Laplacian Eigenmaps Latent Variable Model (LELVM). LELVM is a recently introduced probabilistic dimensionality reduction modelthat combines the advantages of latent variable models--a multimodal probability density for latent and observed variables, and globally differentiable nonlinear mappings for reconstruction and dimensionality reduction--with those of spectral manifold learning methods--no local optima, ability to unfold highly nonlinear manifolds, and good practical scaling to latent spaces of high dimension. LELVMis computationally efficient, simple to learn from sparse training data, and compatible with standard probabilistic trackers such as particle filters. We analyze the performance of a LELVM-based probabilistic sigma point mixture tracker in several real and synthetic human motion sequences and demonstrate that LELVM not only provides sufficient constraints for robust operation in the presence ofmissing, noisy and ambiguous image measurements, but also compares favorably with alternative trackers based on PCA or GPLVM priors. Recent research in reconstructing articulated human motion has focused on methods that can exploit available prior knowledge on typical human poses or motions in an attempt to build more reliable algorithms. The high-dimensionality of human ambient pose space--between 30-60 joint angles or joint positions depending on the desired accuracy level, makes exhaustive search prohibitively expensive. This has negative impact on existing trackers, which are often not sufficiently reliable at reconstructing humanlike poses, self-initializing or recovering from failure. Such difficulties have stimulated research in algorithms and models that reduce the effective working space, either using genericsearch focusing methods (annealing, state space decomposition, covariance scaling) or by exploiting specific problem structure (e.g.