Point cloud registration by discrete spin image and normal alignment radial feature

Abstract

Point cloud registration is a 3D data processing procedure that stitches two or more point clouds together in environment modeling and other related fields. When modeling the plant in the nature environment accurately, the field of view of the scanner is usually so limited that it is hard to acquire the whole point cloud of the plant, so point cloud registration is necessary. The spin image describes the characteristics of point cloud and has great potential in the feature based point cloud registration. In this paper, we propose a registration algorithm based on Discrete Spin Image (DSI) combined with Normal Alignment Radial Feature (NARF), which improves the process of normal calculation and computational efficiency. It is robust under the influence of noise and density of point cloud. Experiments show that the registration speed is increased by at least 6 times and the registration accuracy is about two thirds of average distance of points in point cloud.