Gaussian Belief Propagation-Based Multiview Multiextended Target Tracking With Occlusion

To perform multiview (MV) multiextended target tracking (METT) with occlusion, a Gaussian belief propagation (GaBP)-based MV fusion (GaBP-MVF) algorithm is proposed. A concept of “virtual target” is presented to describe the state of an unobstructed part of the target. The “virtual targets” generate the “partial measurements” affected by occlusions through a spatial measurement model. Subsequently, the closed-form joint posterior probability density function (pdf) of virtual targets is formulated. After factorizing the pdf, a factor graph-based GaBP algorithm is derived for moment estimation of virtual targets’ states. Lastly, sensor-derived estimates are regarded as local estimates and forwarded to a fusion center for updating the global estimate. The virtual measurements are generated by a virtual measurement model (VMM) using the predicted global estimate. Then, the global estimate is updated by minimizing the distance between features extracted from virtual measurements and local estimates. The effectiveness of the proposed algorithm is evaluated in simulation and experiment.