A Study in Object Detection and Classification Performance by Sensing Modality for Autonomous Surface Vessels

<p>This research presents a quantitative performance comparison between light detection and ranging (LiDAR) and vision-based sensing for real-time maritime object detection on autonomous surface vessels. Using Embry-Riddle Aeronautical University’s (ERAU) Minion platform and 2024 Maritime RobotX Challenge data, this study evaluates the detection of six maritime object categories using two representative models. YOLOv8 provides a neural network vision-based method, and GB-CACHE provides a deterministic LiDAR-based method. Both models have been previously demonstrated to run in real time on uncrewed surface vessels (USVs). The evaluation methodology encompasses multi-sensor calibration, real-time performance analysis, and the introduction of a late-fusion strategy in the image frame to reduce bounding-box uncertainty. Performance metrics include training requirements, precision, recall, mean average precision (mAP), and computational efficiency. YOLO achieves low-latency, high-mAP visual detection, while GB-CACHE provides deterministic, CPU-level runtime guarantees with high geometric accuracy, and the Kalman weighted fusion guarantees an improvement in the bounds of the region of interest.</p>

Name thesis:degree_name

Doctor of Philosophy in Mechanical Engineering

Level thesis:degree_level

Dissertation - Open Access

Discipline thesis:degree_discipline

Mechanical Engineering

Year

2025

Author dc:creator

• Lane, Daniel

• Real-Time; Sensor; Fusion; Maritime; Perception
• Robotics

Repository record dc:identifier

https://commons.erau.edu/edt/937

OAI identifier oai:identifier

oai:commons.erau.edu:edt-1992