Tracking in High Target Densities Using a First-Order Multitarget Moment Density
Friday, October 4, 2002 - 10:10am - 1:00pm
Ronald Mahler (Lockheed Martin)
This talk addresses the problem of detecting and tracking large numbers of non-cooperative targets in a cluttered background. The usual approach, which is computationally intractable in general, would be to attempt to detect and track each and every target or potential target. The proposed approach uses the opposite strategy: it attempts to track only what is knowable (initially, geometrical shape and target density) and only later attempting to resolve individual targets out of the multitarget background as (and if) more data becomes available. From a mathematical point of view the approach is novel because the multitarget scenario is modeled as a random measure (specifically, a multidimensional random point process) and the optimal (but intractable) recursive Bayes filter is approximated by propagating the first moment measure (more accurately, its density function) instead of the full multitarget posterior density function.