GM-PHD filter for multiple extended object tracking based on the multiplicative error shape model and network flow labeling

Abstract

In this work, we propose a novel implementation of the Probability Density Hypotheses (PHD) filter for tracking an unknown number of extended objects. For this purpose, we first show how a recently developed Kalman filter-based method for elliptic shape tracking can be embedded into the Gaussian Mixture PHD (GM-PHD) filter framework. Second, we propose a track labeling method based on a Minimum-Cost flow (MCF) formulation, which is inspired by tracking-by-detection algorithms from computer vision. In conjunction with the GM-PHD filter and using a dynamic-programming approach to solve the network flow problem, the overall method is able to achieve a consistent and efficient tracking of multiple extended objects. The benefits of the developed method are illustrated by means of simulated scenarios.