Soft Information Fusion for Tracking Multiple Targets in NLoS Environments

An efficient approximation of the latent soft information (SI) is a key enabler for tracking multiple targets in time-varying non-line-of-sight (NLoS) environments. Such information can be approximated by associating numerous NLoS observations with the targets' movement, a novel paradigm that transitions from the traditional “tracking after localization” approaches to a more comprehensive “tracking with localization” method. In particular, we establish the extended states (ETs) for targets to associate with their possible measurements in multipath scenarios. The joint distribution of ETs is regarded as an approximation of the latent SI at the sensor level. The information carried by ETs is then transferred to the target level for tracking using a variational message passing approach, and subsequently fed back to the sensor level to estimate the SI for future steps. We establish a factor graph depicting the propagation of information from multi-sensor measurements to multi-target states and vice versa. Numerical results including unmanned aerial vehicles (UAVs) validate the superior performance of our proposed algorithm compared to alternative approaches.