Stuart Russell
UC Berkeley
"Tracking Many Objects With Many Sensors"
5/10/2000: [time not recorded]
[location not recorded]
Abstract: Keeping track of multiple objects over time is a problem that
arises in many real-world domains. The problem is often complicated by
noisy sensors and unpredictable dynamics. The data association
literature describes many algorithms that solve special cases or
provide heuristic approaches for various applications. This talk
begins by setting up the general problem as an instance of
probabilistic inference, and points out its intractability. An earlier
method due to Huang and Russell is described and shown not to scale to multiple
sensors. It is shown that estimation of intrinsic object properties
allows scaling through a decomposition of the global likelihood. The
inferenceproblem is solved by a polynomial-time approximation scheme
based on Markov chain Monte Carlo simulation. Experiments with a
freeway traffic simulation suggest that the method allows accurate
estimation of long-range origin/destination information even when the
individual links in the sensor chain are highly unreliable.
If time permits, I will describe an extension of these ideas
to the general problem of inference in first-order probabilistic logics.
Joint work with Hanna Pasula, Mike Ostland, Ya'acov Ritov
Last updated: Mon Jun 19 17:44:06 2006
 |