A Weight Redistributed GM-PHD filter Accounting for Stochastic Missed Detection

Multi-Target tracking is significantly challenging due to the complicities of data association and trajectory correlation. Discontinuous observation sequences evidently cause interruptions on both data association and trajectory correlation, and finally resulting target tracking loss and missed alerts. The Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter is commonly used in multi-target tracking. Under the assumption of constant target detection probability, GM-PHD filter accurately estimates the number of targets and their motion states. However, when the sensor experiences stochastic missed detection of any target member, traditional GM-PHD filter immediately terminates the corresponding trajectory, and subsequently results in target loss and missed alert. To eliminate the risk of missed alerts caused by missed detections, a GM-PHD filter characterized by weight-redistribution is proposed by introducing a dynamic adjustment mechanism on target detection probability, this robust filter guarantees both the estimate accuracy on target number and the tracking stability even stochastic missed detection occurs. Simulation results across multiple scenarios are carried out to demonstrate the significance of the proposed filter.