Stochastic Outlier Selection via GM-CPHD Fusion for Multitarget Tracking Using Sensors With Different Fields of View

In response to the issue of different perspectives affecting fusion accuracy in distributed multisensor multitarget tracking (DMMT), this study investigates the fusion strategies for multitarget tracking with different field-of-view sensors. We explore the impacts of different fields of view (FoVs) on the multisensor geometric average/arithmetic average (GA/AA), generalized covariance intersection (GCI) fusion strategies, and differential perspective boundary segmentation using examples. A Gaussian mixture cardinalized probability hypothesis density (GM-CPHD) fusion method for selecting random outliers in multitarget tracking based on limited perspective sensors is proposed. The boundary is segmented according to different perspectives, and the posterior intensity function is decomposed into multiple sub-intensities using stochastic outlier selection (SOS) clustering. The distribution of the number of targets in the corresponding region is characterized using a multi-Bernoulli reconstruction cardinal distribution. The complexity of the validation algorithm is similar to that of cardinalized probability hypothesis density (CPHD) filtering. A simulation verifies the robustness and effectiveness of this method. By calculating optimal sub-pattern assignment (OSPA) and algorithm runtime, it can be verified that: the algorithm is more robust based on the reduction in detection probability and the increase in the clutter rate.