Research On Color Point Cloud Registration Algorithm

With the development of digital and 3D reconstruction technology,3D point cloud registration has become a research hotspot in reverse engineering and many fields.Point cloud registration is a process that unifies point cloud data collected from different perspectives into the same coordinate system by computing the transformation matrix,the purpose is to obtain the complete point cloud data of the object.At present,most of the registration algorithms are mainly based on the description of the geometric characteristics of point cloud data.However,with the emergence of a new generation of portable and cheap machine vision equipment such as Kinect,which can simultaneously collect 3D coordinates and surface texture information of objects,the registration algorithm has brought new feature information and improvement space.This paper proposes a feature descriptor based on color distribution and fuses it with the geometric feature descriptor FPFH to design a hybrid feature descriptor.The specific content is:(1)In order to improve the efficiency and accuracy of the registration algorithm,a color feature point detection method is designed in this paper.The basic principle is: if there is a certain gap between the gray value of a point and its neighboring points,the point is considered as a feature candidate point.Finally,non-maximum suppression is used to determine the feature points.(2)After deeply studying the local brightness order pattern in the two-dimensional image recognition algorithm,this paper improves and expands it to design a new three-dimensional color point cloud registration algorithm based on the local gray order pattern.The algorithm first encodes neighboring points according to the gray value arrangement around each point in the neighborhood of the key points,and uses the gray difference of the points around the neighboring points to assign weights to their respective feature codes.Then,these feature codes are added and cascaded to construct the feature description vector of the key points.This paper also designs an acceleration method for correspondence matching.This method considers that the gray values between the corresponding point pairs are very similar,so when determining the corresponding point of a point,it is only necessary to search in the points close to its gray value without comparing all the points.This not only increases the matching accuracy,but also reduces the search range of matching points and improves the speed of feature matching.(3)In order to make full use of all kinds of information in the color point cloud data andimprove the registration accuracy of the point cloud,this paper designs a hybrid feature descriptor which combines the color and geometric features.The descriptor combines the geometric features in the fpfh descriptor with the local gray-scale order mode and the color features in the cshot,so that the two features of the point cloud can complement each other to a certain extent.With VC ++ as the development platform,experiments and comparisons are performed on the public RGB-D Object Dataset dataset.The results show that the feature descriptors proposed in this paper can not only complete the registration of color point clouds,but also improve the matching rate and accuracy of registration,and have good robustness.