Point cloud inpainting with normal-based feature matching

With the development of LiDAR technology, point cloud as a data format for representing 3D objects has become more and more widely used. However, negative factors like occlusion or the unfavorable properties of the material surface will lead to the presence of geometric deficiencies, which usually exhibit as holes in point clouds. To solve this kind of problem, point cloud inpainting is proposed. In this paper, we propose an improved point cloud inpainting method, which searches for the proper context for the detected hole and then adopts a more efficient feature matching strategy to refine data source in the hole-filling step by aligning a set of corresponding feature points. Experimental result with comparisons demonstrates its competitive effectiveness with an average 14% gain in GPSNR and better inpainting quality from a subjective perspective.