PlanarSharpen: An Unsupervised Sharp Feature Restoration Method
Yaning Wang, Jianhui Nie · 2024
Sharp features are an important part of point clouds, which contain rich information, but measurement errors and improper data processing can obscure these sharp features. In order to recover and enhance the corrupted sharp features, this paper proposes an unsupervised method to achieve the goal by self-learning to segment the point cloud of the feature region and find the potential planes it contains, and then reconstructing the point cloud and normal vectors at the real location by inferring the location of the real sharp features using the intersecting lines between these potential planes. Experiments show that the method still performs well without any data annotation, has strong generalization ability, and also has good robustness to noise.