An information theoretic approach based Kullback-Leibler discrimination for multiple target tracking

Abstract

To task space-based sensors to efficiently estimate the states of targets, an information theoretic approach is developed based on Kullback-Leibler (KL) discrimination for myopic sensor resource allocation. The technique employs the principle that sensors should take actions that maximize the expected KL discrimination as information gain. Calculate KL discrimination between the priori state probability distribution of targets in cells and the state probability distribution after dummy observations, and use expected KL discrimination to determine the best sensing action to take before actually executing it. Because targets' possible locations and possible dummy observations become a great many with target number and cell number increasing, algorithm modification is designed to combine the states with the same likelihood functions to speedup calculation. Finally the effectiveness of the proposed approach is evaluated by simulations and is verified that it is more effective and accurate to estimate target state and reduce uncertainty than other candidate methods especially in conditions of low SNR or poor sensors.