Likelihood-surface based discretization for tracking via tree search

Abstract

A new discretization technique based on local maxima of the observation likelihood surface is proposed for tree-search based tracking of dim targets in heavy clutter. The joint likelihood of sensor observations over the target state space is evaluated in the vicinity of the previously estimated target state, and its local maxima are selected as new states for discretization. The discretized states are used to build a search tree, which is navigated using the stack algorithm to approximate the maximum a posteriori tracking solution. Simulation results on a benchmark active sonar data set reveal that the proposed algorithm is able to follow dim maneuvering targets without track fragmentation.