Author(s)

Jiaying Lin, Giovanni Campa, Christian-Eike Framing, Jan-Jöran Gehrt, René Zweigel & Dirk Abel

Abstract

Environmental perception and monitoring play essential roles in maritime automation. Besides radar, the use of LiDAR for maritime surveillance has been increasing in recent years thanks to its high accuracy, high data density and good robustness against varying lighting conditions. This paper presents a novel approach for an adaptive shape-fitting technique using LiDAR point clouds in maritime applications, improving the object-tracking performance. The clustered LiDAR point clouds are fitted into bounding boxes or elliptic cylinders depending on their geometric shapes. A fitting score based on mean squared error is used for the shape decision. Afterwards, the extracted objects are associated with those in the past frames and tracked using an adaptive extended Kalman filter. The proposed algorithm is validated in simulation and post-processing using real-world test data. In simulations, the proposed adaptive shape-fitting technique shows a high object positioning and heading accuracy and guarantees a good object-tracking behaviour with a positioning error of 1.5 m. The proposed algorithm’s efficiency and robustness are further validated using test data recorded in the real-world using an unmanned surface vehicle equipped with LiDAR and GNSS in Rostock harbour, Germany. Test results show that the proposed adaptive shape-fitting technique helps the multi-object tracker reach a 2D position error of approximately 2 m with an update rate of 10 Hz, which is sufficient for object tracking in maritime applications. The size accuracy is improved by 10%, and heading accuracy is improved by 16% compared with multi-object tracking approaches only using L-shape fitting.

Keywords

maritime surveillance, multi-object tracking, object detection, shape fitting