Proceedings of CAD’22, Beijing, China, July11-13, 2022, 267-270 © 2022 CAD Solutions, LLC, http://www.cad-conference.net Title: An Efficient Feature Point Extraction Algorithm for Noisy Point Cloud

Introduction Point cloud data is more and more widely used in reverse engineering, cultural relic restoration, architecture and many other fields. . In practice, the obtained data is often non-uniformly sampled and of noises. Feature exaction is a key step for the subsequent processing of point clouds such as matching, segmentation and recognition. How to identify point features for noisy point cloud and improve the efficiency are challenging at present. In previous studies, Nieet al. [2] used a surface smooth shrinkage index (SSI) to measure the degree of surface change, and judged feature points according to the absolute value of SSI of each point. This method has a good anti-noise ability and can extract sharp feature points, but it cannot work for smooth features. In practical applications, we find that the regions of smooth features (such as fillets) have a higher density than other non-featured places, as these interested regions are usually scanned multiple times or the scanning orientation is adjusted to obtain a relatively larger scanning point density at these places. Considering this fact, a combined index of density and SSI is proposed so as to recognize smooth features, and the recognition is also accelerated through octree data structure.