Research On 3D Point Cloud Registration Algorithm Based On Geometric Features

With the development and popularization of 3D point cloud technology,3D point cloud image processing technology has been widely used in many fields.3D point cloud image processing technology has become the core of 3D scanning measurement system,3D reconstruction,reverse engineering,computer vision and robot navigation.Since the 3D scanning equipment cannot obtain all the point cloud images of the object to be scanned at one time,it usually moves the 3D scanning equipment around the target object for multiple scans,and then registers the point clouds obtained from different directions to the common coordinate system to obtain the complete 3D model.This process is called 3D point cloud registration.3D point cloud registration is the most basic and important part of 3D point cloud image processing.It consists of coarse registration and fine registration.The registration effect will directly affect the application of 3D image point cloud in related fields.At present,there are some problems in the mainstream registration algorithms,such as easy to fall into local minimum,poor registration effect in large-angle rotation,slow registration speed and low accuracy in largescale point cloud.Therefore,the research on 3D point cloud registration technology is still important and meaningful.This paper mainly studies the registration method based on geometric features in rough registration,analyzes the deficiencies and defects in the existing algorithms,and optimizes the registration accuracy and registration efficiency mainly.The main work of this paper is as follows:1.The concept of quadric error for surface simplification is introduced into point cloud registration.Taking the quadric error cost as a feature,a registration algorithm based on quadric error cost(Quadric-cost)is proposed.According to the quadratic error cost of the points,the points are sorted and grouped,and the center points of each group are calculated.These points form a feature point group,which is used as a statistical histogram descriptor of the feature description information of the registration algorithm.The advantage of quadratic error is that it describes geometric information,which has relatively stable rotation and affine invariance and can accurately locate the corresponding points through a small amount of calculation.Experimental results show that the proposed algorithm improves the efficiency and accuracy of registration compared with other registration algorithms.2.In order to further improve the registration accuracy and improve the registration effect in large-scale rotation,a registration method based on the combination of high and low cost(Quadric-HLcost)is proposed.The high-cost representation feature,that is,the high-cost point of the same model is relatively fixed,and will not change greatly with the simplification of the model.The bounding box of the low-cost point will not change greatly,and its center point is also stable.By calculating the weight values of high-cost points and low-cost points,a group of feature points is formed,and then the histogram descriptor is calculated.The proposed registration algorithm outperforms popular algorithms in large-scale rotation and registration efficiency and accuracy.To sum up,the registration of 3D point clouds is studied in this paper,and two registration methods Quadric-cost and Quadric-HLcost are proposed.The experimental results show that these methods can effectively improve the efficiency and accuracy of registration.