Gaussian Mixture Probability Hypothesis Density Filter with State-Dependent Probabilities
Yi‐Chieh Sun, Inseok Hwang · 2021 European Control Conference (ECC) · 2021
The Gaussian mixture probability hypothesis density (GM-PHD) filter has been successfully applied to various multiple target tracking (MTT) applications due to its ability to estimate both the number of targets and target states from noisy measurements effectively and efficiently. However, since the GM-PHD filter assumed the target detection and survival probabilities and the birth rate to be constant, its performance could be degraded when the targets are temporarily occluded or when measurements are missing. To address this, we propose the state-dependent GM-PHD filter which explicitly considers the state-dependent probability of detection and survival as well as the state-dependent birth rate. In addition, in order to consider the different kinds of maneuvers of the target, we use a hybrid system model for a target. The performance of the proposed algorithm is illustrated with an example in a road traffic simulation, which includes target occlusion, and compared with that of the original GM-PHD filter.