An Improved Peak Extraction Algorithm for Probability Hypothesis Density Particle Filter

摘要

The Probability hypothesis density (PHD) particle filter is a new practical method to solve the unknown time-varying multi-target tracking problem. Peak extraction method is needed to detect the target states from the posterior PHD approximated by the particles and their weights. Tobias' peak extraction algorithm sequentially removes the PHD component of each target, which works more efficiently than the k-means clustering and expectation-maximum algorithm. However, it becomes unreliable in dense targets environment. This paper improves Tobias' peak extraction method in finding the particles representing a target and removing the effect of the target peaks. The proposed algorithm exploits a layer labeled technique to find the PHD component for a single target and segment the weights of these particles according to their corresponding distances to the candidate of the target state. Demonstrations show that the proposed algorithm is more accurate and efficient compared with current-used peak extraction algorithms.