Abstract
The cardinalized probability hypothesis density (CPHD) filter is an alternative approximation to the full multi-target Bayesian filter for tracking multiple targets. However, although the joint propagation of the posterior intensity and cardinality distribution in its recursion allows more reliable estimates of the target number than the PHD filter, the CPHD filter suffers from the spooky effect where there exists arbitrary PHD mass shifting in the presence of missed detections. To address this issue in the Gaussian mixture (GM) implementation of the CPHD filter, this paper presents an improved GM-CPHD filter, which incorporates a weight redistribution scheme into the filtering process to modify the updated weights of the Gaussian components when missed detections occur. In addition, an efficient gating strategy that can adaptively adjust the gate sizes according to the number of missed detections of each Gaussian component is also presented to further improve the computational efficiency of the proposed filter. Simulation results demonstrate that the proposed method offers favorable performance in terms of both estimation accuracy and robustness to clutter and detection uncertainty over the existing methods.
Highlights
Multiple targets tracking (MTT) is a key technology for many practical applications in both military and civil fields [1,2]
We propose an improved Gaussian mixture (GM)-cardinalized probability hypothesis density (PHD) (CPHD) filter, which aims at addressing the aforementioned drawbacks of the original version
We compare the performance of the proposed GM-CPHD filter without gating, In this part, we filter, compare performance of GM-CPHD
Summary
Multiple targets tracking (MTT) is a key technology for many practical applications in both military and civil fields [1,2]. The MTT algorithm need to jointly estimate the time-varying number of targets and their individual states via using the measurements corrupted by noise and clutter. The PHD filter propagates the posterior intensity of the multi-target state, while the CPHD filter propagates the cardinality distribution, i.e., the probability distribution of the number of targets. The unknown multi-target state generates corresponding measurements whose number is time-varying at each time step. The random finite set (RFS) approach provides a mathematically-elegant treatment of multi-target systems by modeling the collections of target states and measurements as RFSs. For example, if there are nk targets with states xk, , xk,2 , . Zk,mk at time k, the RFS representation of the multi-target state and measurements are respectively defined as [7,11]: Xk = xk, , xk,2 , . Some clutter measurements may be collected, and some of the existing and newborn targets may not be detected due to the imperfect detectors
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have