Semantic Map-based Localization of USV Using LiDAR in Berthing and Departing Scene

High-precision localization is one of the most critical components of unmanned surface vehicle (USV). However, the accuracy and robustness of traditional GPS-based localization methods in complex berthing and departing scenarios are still limited. This paper proposes a novel localization method of USV based on LiDAR semantic and geometric information in the berthing and departing scene, which integrates semantic information such as shoreline, ships, and buildings into the SLAM system to enhance the perception ability and localization accuracy of USV. First, a CNN-based berthing and departing scene perception method using LiDAR is presented, which make USV get semantic information from the port accurately. And then, to eliminate the impact of dynamic objects on registration and mapping, an enhanced semantic odometry of USV is adopted, and semantic information can be used to improve robustness while maintaining efficiency. Then, based on the factor graph, semantic LiDAR odometry factors and IMU (inertial measurement unit) pre-integration factors are constructed to optimize the pose and semantic map. Finally, experiments are conducted on real-world berthing scenarios to demonstrate the effectiveness of the proposed method. Especially, even in the case of GPS signal loss, robust and accurate localization of USV in berthing and departing scene can still be obtained.