Abstract: When using the FPFH feature of point cloud for 3D object recognition or registration,FPFH feature descriptor is arbitrarily and inefficiently calculated by subjectively adjusting the neighborhood radius,and the whole process can not be completed automatically.This paper proposed an adaptive neighborhood-selection FPFH point cloud feature extraction algorithm to solve this problem.Firstly,the point cloud densities of many pairs of point clouds were estimated.Secondly,the neighborhood radii were computed to extract the FPFH features for SAC-IA,and the radii and the densities were counted when the registration performance is the most optimal,and then the Cubic Spline Interpolation Fitting was used to fit the function expression of the radii and the densities to form the adaptive neighborhood-selection FPFH feature extraction algorithm.The experimental results show that this algorithm can adaptively choose the appropriate neighborhood radius according to the density of point cloud,improves the FPFH feature matching perfor-mance,and improves the computing speed at the same time,indicating that the proposed algorithm is of important guiding significance.

Key words: Fast point feature histograms, Neighborhood radius, Point cloud density, Sample consensus initial alignment