A Real-Time Fusion of Two-Stage Point Cloud Clustering and Saliency Image for Water Surface Object Detection

Unmanned surface vessels may encounter unknown surface obstacles when sailing. Accurate detection has a significant impact on the subsequent decision-making process. In order to deal with the complex water environment, this paper proposes an object detection framework based on the fusion of LiDAR and camera. The detection framework can achieve real-time and accurate water surface object detection without training, and has strong anti-interference ability. The detection framework achieves the data fusion of LiDAR and camera through external calibration and then uses the detection algorithm of sky–sea boundary (SSB) to establish a clear search area for LiDAR. Then, a two-stage clustering algorithm based on point cloud attributes and distribution information achieves more accurate segmentation. The region of interest (RoI) is obtained from the detection results by image projection. Finally, the region of interest is finely segmented by the saliency object detection algorithm. The experimental results show the effectiveness and robustness of the algorithm.