Relative Position Estimation Algorithm for Inspection UAV Based on Point Cloud Registration

As a key structure of contact networks, efficient operation and maintenance of contact network support device is crucial to the safety of railway operations. Due to the specificity of the location of the contact network support device, it needs to rely on a specific shooting angle and distance, which brings certain difficulties for the manual over-the-horizon control of the UAV. Therefore, in order to assist the autonomous data acquisition of railway patrol UAV, a method based on point cloud registration is proposed to solve the optimal relative shooting position. By learning the features between the preset template and the actual point cloud, and combining with SVD to solve the rotation translation matrix, the method can guide the UAV to the designated location. In the experimental verification phase, the point cloud registration algorithm is trained using 3DMatch data set, and then tested using the laboratory data set that is completely unknown to the training model. Experimental results demonstrate that the angular errors along all axes are generally within 1°, with translational errors mostly below 5 cm. To further validate the feasibility of the proposed method, visualization analysis was conducted using the WHU railway dataset. The results indicate that the method maintains excellent registration accuracy in real-world scenarios, meeting the precision requirements for positioning in railway catenary inspection tasks.