Point Cloud Registration Based On RGB-D Data

Point cloud data is a set of discrete points that approximately represent thegeometry of objects(such as location,color information,etc.The acquisition of point cloud data can be achieved through 3D scanner equipments(such as Kinect).Point cloud registration is the process of merging the point cloud data collected from different perspectives to the same coordinate system,the ultimate goal is to obtain the complete and consistent surface description by point cloud.Point cloud registration is an indispensable part of automatic 3D reconstruction,and is also the basis of virtual reality applications.The point cloud registration has already been widely used in the field of human-computer interaction,robotics,reverse engineering and other fields as well.Point cloud registration can be categorized into coarse registration and fine registration,this paper focuses on the two aspects,and proposes the strategy of point cloud registration with the property of fast convergence and high precision.The main contents of this thesis are as follows:To attack the problem of filter data and the target extraction,by experimenting on various filtering algorithms,it shows that the multi frame mean filter is suitable for Kinect depth image(depth data and point cloud data filtering algorithmconverts).Furthermore,based on the threshold segmentation and RANSAC algorithm,this thesis proposes a target point cloud extraction method based on Kinect optimal horizon.In this thesis,the limitations of ICP algorithm are analyzed and experimented.It shows that the offset of the initial position is the most important factor that affects the accuracy of the ICP algorithm.The ICP algorithm is accelerated by the Kd-tree nearest neighbor search and the voxel grid sampling strategy.As for the coarse registration,by taking into consideration of the spatial location and color features of the key points in the field,this thesis proposes a color key point descriptor which is suitable for XYZRGB point cloud data.Combining with the RANSAC algorithm,this thesis propose a joint coarse registration algorithm,which takes into account both the accuracy and efficiency of the coarse registration.Finally,this thesis presents a more comprehensive point cloud registration strategy,which includes the acquisition and preprocessing of point cloud,coarse registration and accurate registration.And through experiments,The feasibility of this strategy under different scenarios are verified through experiments.