Research On Point Cloud Registration Technology Based On Color Feature

With the improvement of point cloud acquisition technology and the large number of low-cost point cloud acquisition devices led by Microsoft's Kinect,3D reconstruction technology is widely used in many fields such as reverse engineering,machine vision,virtual reality,and human-computer interaction.Point cloud registration technology is an important part of 3D reconstruction technology.Many scholars have improved the registration algorithm.However,for scattered point clouds without any predictive information,the existing iterative closest point(ICP)algorithm is difficult to achieve a good balance between registration speed and registration accuracy.In order to solve this problem,this paper proposes a color point cloud initial registration algorithm based on key points and an ICP point cloud accurate registration algorithm based on color features.The specific research work of this paper is as follows.Firstly,aiming at the problem that the existing initial registration algorithm is too slow or too low precision in registering scattered point clouds,a color point cloud initial registration algorithm based on key points is proposed.The algorithm uses voxelized grid downsampling to extract key point clouds.By extracting key points from two points to be registered,the number of participating points in the registration process is reduced,and the registration speed is improved.In the process of searching corresponding point pairs,color features are introduced,combining with fast point feature histogram(FPFH)to match corresponding point pairs.The random sampling consistency(RANSAC)algorithm is used to remove the error point pairs and improve the point cloud registration accuracy.Then,aiming at the slow speed of registration of scattered point clouds with existing ICP algorithms,an accurate ICP registration algorithm based on color feature is proposed.The algorithm uses k-tree to speed up the search for corresponding pairs of points,uses the color features of the point cloud to establish constraints,sets the threshold to remove the wrong point pairs,and ensures the registration accuracy while improving the registration speed.Finally,the running results of the two registration algorithms are demonstrated byexperiments,and then the effectiveness of the algorithm is verified from the running time and registration error of the algorithm.The experimental results show that the proposed algorithm improves both registration speed and registration accuracy.