Sequential Monte Carlo Methods

I am interested in the use of sequential Monte Carlo methods and particle filters for state estimation in robotics and vision. In joint work with Dieter Fox, Wolfram Burgard, and Sebastian Thrun we saw the potential of particle filters in the context of mobile robot localization, which led to the development of the highly popular Monte Carlo Localization algorithm. I am currently investigating, with my students, novel sequential Monte Carlo methods that are applicable in domains where traditional particle filters fail. For example, our recent work on Rao-Blackwellized EigenTracking makes particle filters cope with complex, subspace-based appearance representations as needed for complex visual tracking tasks. Last but not least, the recently developed MCMC-based particle filter, which replaces the traditional importance sampling step with the much more efficient MCMC sampler, promises to be a leap forward in the tracking of many interacting targets. A lot of this work is done in the context of the BioTracking project, an NSF-funded effort that is lead by Tucker Balch and myself.


Relevant Publications

Related Links

Frank Dellaert Home
Computational Perception Lab

Frank Dellaert's Homepage