Motion Constrained Point Cloud Matching for Maritime Tracking

Extended Object Tracking (EOT) plays a critical role in accurately estimating the state of nearby vessels in confined regions, such as urban waterways. By estimating a vessel’s extent, it is possible to improve velocity estimation and motion prediction. Traditional EOT methods in the maritime domain that use point clouds from Light Detection And Ranging (LiDAR) commonly parameterize target extents. However, these approaches often struggle with data association and clutter removal, particularly when using 2D parameterizations. To address these limitations, we propose a 3D EOT framework leveraging motion-constrained Iterative Closest Point (ICP) displacements integrated into an Error State Kalman Filter (ESKF). Our approach incorporates maritime-specific motion constraints, enabling robust target tracking even with sparse measurements. Evaluation on real-world maritime data demonstrates that the proposed 3D tracker significantly outperforms a 2D Gaussian Process Extended Object Tracking (GP-EOT) tracker, offering enhanced resilience against wake-induced noise and delivering superior state and extent estimation accuracy.