摘要
Abstract
Aiming at the low computational efficiency,noise sensitivity and instability of traditional key point extraction algorithm of 3D laser scanning point cloud,we proposed a key point extraction algorithm based on feature space value screening.Firstly,the algorithm establishes a network model for feature extraction.The input of model is the set of angles between the normal vectors of all points in the point cloud and their neighborhood points,and the output of model is one-dimensional feature space vectors.Then,this algorithm sorts the feature space vectors ac-cording to the value size,takes the corresponding points with the maximum n values as the key points.The experimental results show that the key points of algorithm have high repeatability and operation efficiency,and have better robustness to noise.The average repetition rate in small scene noise data is increased by about 23.0%,and in large scene data is increased by about 9.3%.The key points are mainly distributed in the area with obvious feature changes,which can better express the feature of point cloud.关键词
三维激光扫描/关键点检测/法向量夹角/特征空间/重复性Key words
3D laser scanning/key point detection/normal vector angle/feature space/repeatability分类
天文与地球科学
