Automatic Sharp Feature Extraction from Point Clouds with Optimal Neighborhood Size

A novel algorithm is proposed for extracting sharp features automatically and at optimal scale from point clouds. First, the vector between a given point and the centroid of its neighborhood at a given scale is projected on the normal at this point. This projection is called the ’projected distance’ at this point. The projected distance and surface normal vector are recalculated at several scales for each point. In a second stage, the projected distance at different scales is analyzed in order to choose the optimal neighborhood size and update the final projected distance value for the point. Finally, Otsu’s method is applied to the histogram of the final projected distances on the cloud in order to select the optimal threshold value which determines whether points are on a sharp feature or not. The method has many advantages such as automatic selection of threshold, optimal neighborhood selection, accurate and robust detection of sharp features on a wide variety of objects. To demonstrate the robustness of the method, it is applied on both synthetic and complex point clouds with different noise levels.