A model-based, generative and stochastic method for human motion capture using hierarchical particle filters
Abstract
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT REQUEST OF AUTHOR.] Searching in probability spaces can prove to be an impractical task due to the high- dimensionality of the state vector. In the context of tracking human pose through image features in video sequences, the number of degree-of-freedom (DOFs) of the human body forces the search to be done using an exponentially large number of possible configuration states. In this dissertation, we stated that the computational complexity of this search can be greatly reduced by the introduction of a hierarchical model for the propagation of the state variable and by the efficient selection and synthesis of configuration states through this hierarchy. We demonstrated this claim by developing a new hierarchical framework for tracking human pose. Extensive experiments on a public benchmark dataset demonstrate comparable tracking errors to the state-of-the-art and up to 60% computational reduction.
Degree
Ph. D.
Thesis Department
Rights
Access is limited to the campuses of the University of Missouri.