Point set registration-based image stitching in unmanned aerial vehicle transmission line inspection

To address the limitations of narrow field of view and insufficient target information in single frame images obtained by airborne cameras, this paper proposed a novel image stitching method using an improved robust point set registration algorithm based on neighborhood and local structural features(PR-NLS). The proposed method combines stable local and global constraints between two corresponding point sets, weighted spatial distance is introduced to obtain the local neighborhood of the motion vector to reduce the complexity of the point set, and combines weighted spatial distance with feature descriptors to improve the model fit. Then, the PR-NLS algorithm is added to the image stitching to stitch the local images of the power pole tower and corridor collected by the Unmanned aerial vehicles (UAV) power inspection. Experiments show that the image splicing algorithm based on PR-NLS can realize local image splicing, and the matching accuracy can reach 98% Compared with other classical algorithms, it has less mismatching points and higher image average entropy, and solves the problem of incomplete image information taken during the power inspection of transmission line towers and corridors.