Surface Target Detection Of Unmanned Surface Vehicle Based On LIDAR Data

With the rapid development of computer technology and communication technology,the Unmanned Surface Vehicle(USV),as the surface unmanned carrier of the maritime transportation system,is not only widely used in military activities such as reconnaissance,jamming,mine sweeping and assault,but also plays an important role in civil fields such as surface cruise,maritime search and rescue,environmental monitoring and resource exploration.The autonomous obstacle avoidance navigation of unmanned boats requires real-time perception of surrounding environmental information,in order to detect obstacles around unmanned boats.However,Li DAR plays an important role in detecting short range targets in unmanned boats due to its high ranging accuracy and strong anti-interference ability.In view of the problem that the existing public detection algorithm is applied to the high rate of missed detection under the ship sway state,this thesis combines the positioning and position analysis information of integrated navigation GNSS/IMU to achieve the distortion removal of R-Fans-16 laser radar data and the stable detection under the ship sway state.The thesis will focus on the following three parts.(1)In view of the large amount of original lidar data,Voxel Grid filter is used to realize point cloud de-sampling.The filtered 3D point cloud data is compressed by two-dimensional grid map projection of adaptive morphology.By analyzing the characteristics of the wake flow point cloud and combining with the idea of Ray Ground Filter algorithm,the RANSAC algorithm is improved to realize the dynamic adjustment of the threshold range of the wake flow occurrence area,thus improving the filtering effect of the wake flow point cloud.(2)In view of the large volume of ships on the sea,the traditional detection algorithm is easy to lead to the low efficiency of algorithm implementation caused by the over-segmentation of detection targets and the different sparsity of point cloud density with different distances,this thesis proposes a target clustering algorithm suitable for different distances based on DBSCAN algorithm.Hierarchical clustering is carried out for the objects that are separated from the threshold boundary to avoid over-segmentation of large obstacles at the segmentation boundary.Experiments show that the algorithm proposed in this thesis can achieve efficient and accurate detection results in detecting targets of different sizes at different distances.(3)Based on the unmanned boat equipped with Robot Operation System(ROS),the port environmental target detection is realized.By extracting the key points of the environment point cloud,the calibrated integrated navigation GNSS/IMU is used to associate the key point data of the point cloud between frames.Then the roll,pitch and yaw angles of the lidar are obtained for accurate inter-frame projection of the two adjacent point clouds.The key point association attributes of the two frame point clouds in the projection results are taken as the input of the Kalman filter algorithm,which realizes the dynamic and static attribute judgment of obstacles and the position estimation of the key points of the next frame point cloud.The laser radar detection algorithm proposed in this article efficiently reduces the complexity of the algorithm while meeting the accuracy of target detection.On the one hand,it effectively reduces the clustering over segmentation rate of large targets such as ships and shorelines.On the other hand,under the ROS framework,it ensures the timeliness of subsequent fusion of monocular cameras and depth cameras for effective perception.