An estimation algorithm for improved maritime obstacle detection

This paper presents an advanced estimation algorithm for improving obstacle detection in maritime navigation by integrating data from Velodyne LiDAR HDL-32E and a ZED2I depth camera on the Riptide Unmanned Surface Vehicle. Utilizing a pre-trained YOLOv8L model, the ZED2I camera identifies objects and estimates distances using RGB and depth data. The LiDAR system, developed in Python and C++, features adaptive clustering and an obstacle detector, aligning its point cloud to a 110-degree field of view to match the camera's perspective. A specialized ROS node facilitates data fusion by prioritizing ZED2I camera data, enhancing detection accuracy through comparisons with LiDAR data. Extensive testing, including calibration adjustments like Z-axis alignment and a 4 cm offset correction between the camera and LiDAR, has demonstrated the system's refined accuracy in controlled tests, underscoring the necessity of continuous calibration. This approach exemplifies the benefits of hybrid data management and precise sensor alignment in enhancing maritime navigation safety.