Multi-Sensor Multi-Target Tracking for Maritime Surveillance with Autonomous Surface Vehicles Using Belief Propagation

We present a distributed multi-sensor multi-target tracking algorithm for maritime surveillance using unmanned surface vehicles (USVs) in multi-agent settings. Our approach fuses measurements from radar, Automatic Identification System (AIS), and camera sensors within a Bayesian framework, employing an adaptive particle filtering strategy to jointly estimate the kinematic states and identities of vessels. Our solution incorporates a factorized data association model that integrates cooperative self-reports from AIS with radar and camera measurements, with visual re-identification capability. We evaluate our method using a high-fidelity simulation environment, which generates photorealistic maritime scenarios. Our performance analysis indicates that the integration of camera-based cues improves both the spatial localization and identity consistency, particularly in scenarios with low radar detection probability and non-cooperative targets. Furthermore, the distributed inference framework scales well with the number of USVs, making it well suited for large-scale multi-agent applications. Overall, our work demonstrates that fusing heterogeneous sensor modalities using belief propagation can enhance multi-target tracking performance in congested maritime environments.