基于激光点云的输电线路无人机仿线飞行方法研究OA

An end-to-end method is proposed to address the challenges of stable orientation determination and complete segmentation in UAV line-following flight,where sparse,local and structurally incomplete point clouds make it difficult to extract conductors within real-time sliding windows,thereby affecting planning and control.The local point cloud is projected onto the horizontal plane,and principal component analysis(PCA)in multiple regions is conducted to extract principal axes,and the weighted voting based on linearity is conducted for the robust estimation of the main corridor direction.The orthogonal cross-section is constructed according to the main direction to project the point cloud onto the cross-sections,and conduct the adaptive clustering with density-based spatial clustering of applications with noise(DBSCAN),so as to realize high-integrity segmentation of the point-cloud cluster for a single conductor.Each connected cluster is then fitted with a parabolic model,and the outlier rejection and the raw inlier count are used to select a reliable reference trajectory.The thrust and attitude feed-forward based on the differential flatness of UAVs is designed to realize closed-loop tracking combined with PID controller.The actual radar measurement data and Gazebo simulation results show that,in comparison with traditional method,the proposed method can realize significant improvements in both direction recognition accuracy and segmentation completeness on sparse dynamic point clouds.With the introduction of the differential-flatness feedforward and PID controller,it yields smaller tracking errors and better stability,which can support UAVs to realize stable line-imitating flight.