With the continuous development of water transportation and shipping, the detection and tracking of ships traveling on the river have become increasingly critical. While the image-based ship detection and tracking methods have matured, their inability to directly obtain 3D dimensions and spatial positions due to the lack of depth information remains a limitation. The point cloud data generated by LiDAR, on the other hand, naturally carries precise geometric and distance information, demonstrating significant potential for enhancing ship detection and tracking capabilities. Current 3D point cloud object detection approaches can be categorized into methods based on classical algorithms and those leveraging deep learning. However, classical point cloud algorithms applied to ship detection suffer from poor generalization and difficulty in distinguishing adjacent ship point clouds. To address these challenges, an improved algorithm of PV-RCNN++ based on focal sparse convolution is proposed for ship detection in river channels. The improved algorithm not only effectively differentiates ship point clouds in various situations, but also improves the recognition ability of distant ships, achieving an 11.56% increase in detection accuracy in practical scenarios compared to classical methods. On this basis, a multi-target ship matching and tracking method based on the degree of correlation between ship positions and 3D dimensions is proposed, in which the ICP alignment is used to calculate the ship speed and predict the ship position. Experimental validation results demonstrate that the proposed tracking method exhibits stable performance and achieves accurate ship matching between consecutive data frames.