Near-Shore Mapping for Detection and Tracking of Vessels

TL;DR

This work tackles the problem of detecting and tracking vessels near a dock for autonomous surface vessels, where traditional land masking is insufficient for close-to-shore objects. It introduces an offline LiDAR-based mapping pipeline that integrates camera-derived potentially moving object segmentation to produce a precise 2D map of the docking area, effectively filtering out static land while isolating dynamic targets. A 2D tracker (VJIPDA) operates on detections derived from the mapped scene, benefiting from a full field of view during tracking. The results on real-world near-dock sequences show that accurate maps enable earlier target detection and reduced false tracks, with practical impact for safer docking and harbor operations, albeit requiring reliable vessel detectors and adequate region coverage.

Abstract

For an autonomous surface vessel (ASV) to dock, it must track other vessels close to the docking area. Kayaks present a particular challenge due to their proximity to the dock and relatively small size. Maritime target tracking has typically employed land masking to filter out land and the dock. However, imprecise land masking makes it difficult to track close-to-dock objects. Our approach uses Light Detection And Ranging (LiDAR) data and maps the docking area offline. The precise 3D measurements allow for precise map creation. However, the mapping could result in static, yet potentially moving, objects being mapped. We detect and filter out potentially moving objects from the LiDAR data by utilizing image data. The visual vessel detection and segmentation method is a neural network that is trained on our labeled data. Close-to-shore tracking improves with an accurate map and is demonstrated on a recently gathered real-world dataset. The dataset contains multiple sequences of a kayak and a day cruiser moving close to the dock, in a collision path with an autonomous ferry prototype.

Figures (9)

• Figure 1: Overlain in cyan is an example of land masking. Overlain in orange is a visual and -based map, details are in \ref{['sec:mapping']}. The orange map is more detailed, without potentially moving objects, and gives earlier target detections than the cyan map. A photo of a relevant docking area is in the background. Base image courtesy of Google Maps: Imagery @2024 Airbus, CNES / Airbus, Maxar Technologies, Map data @2024.
• Figure 2: The kayak is undocking and moving into the canal. The approximate path it takes is marked in blue, and the approximate path for the day cruiser is marked in orange. A precise map enables earlier tracking of the kayak compared to an imprecise map, where the kayak can be differentiated from the dock.
• Figure 3: The static floating dock is mistakenly detected as two boats. This needs to be handled to map the dock.
• Figure 4: Examples of annotations that differentiate boats from floating docks. This labeling is used to train the segmentation method.
• Figure 5: What a single frame can look like during mapping. The boat detections are in light orange, the image is brightened in the background and the projected points are in blue. This illustrates how the point selection is done, differentiating between static objects and vessels.