Wake-Aware LiDAR Perception for Unmanned Surface Vehicles via Physics-Informed Filtering and Adaptive Clustering

Accurate environmental perception is a key challenge for unmanned surface vehicles (USVs) in dynamic maritime environments, where waves, wakes, and variable visibility often reduce the reliability of conventional sensing systems. This paper presents a LiDAR-based perception framework that integrates wake-aware filtering and real-time adaptive clustering for maritime object detection and tracking. The framework consists of a multi-frame spatiotemporal fusion module, an Adaptive Dynamic RANSAC (AD-RANSAC) filter, and a Real-Time Adaptive Spectral DBSCAN (RTA-DBSCAN) clustering algorithm. AD-RANSAC combines a Gaussian Mixture Model with a physics-informed wake model to suppress structured and unstructured maritime noise, while RTA-DBSCAN adaptively adjusts the clustering radius according to sensor geometry and local point-cloud density. Extensive sea-trial results show that the proposed method achieves a true positive rate of 93.37%, reduces false positives, and maintains real-time performance, outperforming representative DBSCAN variants and several deep learning-based detectors. The results demonstrate the effectiveness of the proposed framework for robust maritime perception under varying sea states.