Improved GM-PHD Filter for Multi-target Tracking with Dense Clutter and Low Detection Probability

In scenarios with high clutter intensity, Gaussian mixture probability hypothesis density filtering (GM-PHD) based on the random finite set theory offers distinct advantages compared to traditional data association based methods. However, achieving real-time and accurate tracking is highly challenging due to the presence of dense clutter and low detection probability. To address these issues, we proposed an improved Gaussian mixture probability hypothesis density filter (I-GM-PHD). The proposed filter pre-prunes the measurement by each Gaussian component, and adaptively adjusts the clutter intensity parameters in the update process according to the elliptical gate area of the Gaussian component to correct the Gaussian component weight. Simulation results indicate that the proposed I-GM-PHD filter enhances the accuracy of Optimal Subpattern Assignment distance and reduce the standard deviation between the estimated target number and the real target number compared with the GM-PHD filter across various clutter concentrations and detection probabilities.