Automatic UAV Inspection of Hydropower Station Diversion Pipelines Using Robust Cylinder-Like Fitting for Incomplete Point Clouds
Quanxi Zhan, Junrui Zhang, Chenyang Sun, Fenghe Guo, Xinyi Zhao, Chen Gao, Linchuan Yang, Runjie Shen, Bin He · 2024
This study presents a solution for UAV navigation within hydropower station diversion pipelines, introducing a robust cylinder-like fitting algorithm for incomplete point clouds (RCFIC) alongside a UAV state estimation algorithm. The RCFIC algorithm efficiently fits incomplete point clouds to cylinder-like structures using 3D lidar and IMU sensors. It preprocesses the point cloud data, slices it equidistantly along the centerline of the previous frame, and applies the Levenberg-Marquardt algorithm to fit ellipses to the cutting plane points. A line or curve is then fitted to the centers of these ellipses. Using this cylinder-like model and various sensors, the UAV's target point, speed, and yaw are estimated based on geometric relationships. Field tests conducted at a hydropower station validated the algorithm. In both horizontal and oblique straight segments, the RCFIC algorithm achieved over 90% fitting accuracy, while it successfully maintained alignment with the pipeline's centerline in curved segments, demonstrating the method's robustness and precision. This approach enables reliable, automated inspection of diversion pipelines.