An Adaptive Multi-Bernoulli Filter for Coexisting Point Target and Extended Target Tracking With Unknown Detection Probability

This paper proposes an adaptive multi-Bernoulli (MB) filter for coexisting point target and extended target tracking, where the detection probability of each target is unknown and time-varying. The detection probability of the target directly affects the tracking accuracy in multi-sensor multi-target tracking. In realistic tracking circumstances, the detection probability of the target is unknown. Therefore, it is crucial to adaptively estimate the detection probability of the target in real-time. In addition, as the resolution of sensor increases, the point target and extended target often coexist in the field of view. Therefore, it is necessary to construct a generalized measurement model for tracking coexisting point target and extended target. For the coexistence of point and extended target, we model the spatial probability density function which can accommodate both point target and extended target based on a MB filter. For the unknown detection probability, we integrate the detection probability into the target state and the detection probability is characterized as a Beta distribution. To obtain a closed-form solution, the spatial probability density function of each MB component is modeled as a mixture of the Beta-Gaussian distribution and the Beta Gamma Gaussian Inverse Wishart (BGGIW) distribution. Based on the estimated probability of the point target, the target is determined as either a point target or an extended target. Finally, the detection probability, kinematic state, Poisson measurement rate (MR) and extended state of each target are estimated synchronously. Simulation results verify the effectiveness and robustness of the proposed algorithm.