Combining Occupancy Grid Mapping and Extended Object Tracking With the Poisson Multi-Bernoulli Mixture Filter

For autonomous agents to operate effectively, an accurate understanding of both the static and dynamic components of the environment is crucial. A key challenge lies in the correct association of measurements to either the static environment or dynamic objects, as errors in data association can undermine both tracking and mapping. These problems are made even more complex with high-resolution sensors such as LiDARs. This article introduces an extended object Poisson multi-Bernoulli mixture (PMBM) filter that simultaneously tracks extended dynamic objects and maps the static environment. Dynamic objects are modeled using the Gaussian process extent model, while the static environment is modeled using an occupancy grid. Using a joint PMBM density, the filter addresses data association challenges and models undetected objects, providing an elegant solution for track initiation. The occlusion of different components is also addressed. Furthermore, a grid-based representation of the birth and undetected object density is used, aligned with the discretization used for the occupancy grid map. This filter is applied to object tracking and mapping for an uncrewed surface vessel (USV) equipped with a LiDAR navigating in a confined waterway. To mitigate the issue of falsely detected dynamic objects associated with measurements generated by the static environment, we restrict areas for potential new dynamic objects, using prior map information from an electronic navigational chart. Simulation and real-world data demonstrate the method’s capability to track an unknown number of dynamic objects and accurately estimate their extent while mapping the static environment.