Generation,Registration And Fusion Of Multi-modal 3D Point Cloud Data Model

It is difficult to obtain the whole three-dimensional point cloud data of the object because of the limited scope of vision of the coordinate measuring device and the size of the measured object.Due to the collection or processing of point cloud data,there are inevitably different modes of point cloud data in a project.Therefore,registration and fusion of point cloud data obtained under different modes can ensure the integrity of spatial information of collected objects and can be applied to practical projects better.In this thesis,in order to realize the multiple model of point cloud registration and fusion as the goal,the initial research on 3D point cloud registration algorithm is proposed based on the center of gravity and center of mass of rotation invariance initial registration algorithm,by introducing the ICP algorithm and NDT algorithm,to overcome the error under the condition of feature points is more difficult to achieve the initial point cloud registration,finishing more model of point cloud data registration and integration.In terms of improving the accuracy of three-dimensional point cloud acquired by Unmanned Aerial Vehicle(UAV),based on SFM 3d reconstruction technology,the accuracy of dense point cloud can be finally improved by studying image Super resolution restoration(SR)technology.The specific research content includes the following aspects:1.Research on initial 3D point cloud registration algorithmThe initial point cloud registration algorithm is studied.Aiming at the problem that accurate registration algorithms such as ICP in 3D point cloud registration are prone to fall into local optimality,an initial registration algorithm based on geometric barycenter and centroid transformation is proposed.Moreover,the algorithm proposed in this thesis has a low complexity and can complete the initial registration of large-scale point cloud data sets.Firstly,the point cloud is filtered.Then,the initial model of rotation transformation between point cloud data is established by using the center of gravity and center of mass of point cloud data.Secondly,according to the relationship between rotation Angle and registration error,an iterative rotation model is established to find the optimal rotation Angle,and then the initial registration is completed.Finally,ICP algorithm is combined to further accurate registration.2.Improve the accuracy of obtaining three-dimensional point cloud data model based on UAVBased on the SFM technology,3D reconstruction image sequence using chaos,scene 3D reconstruction,in the reconstruction process is extracted from the image sparse point,considering the model of three-dimensional point cloud data based on UAV for low accuracy,through the analysis of computer visual image point cloud to build the main theory and key technology,mainly including the SIFT feature extraction,RANSAC feature matching and dense reconstruction,after matching,diffusion,filter generate dense point cloud with real color,proposed by super-resolution processing the original image data to increase the number of feature extraction,And the method through experiment verification,the results show that after super-resolution processing image data,can increase the sparse point cloud data,from the effect of the reconstruction of the dense point cloud data quality has improved significantly,the experiment indicates that the part can be used to improve in the future based on UAV to obtain the accuracy of three-dimensional point cloud data model.3.Multi-mode point cloud registration and fusionIn this thesis,the experimental data obtained under the two modes are large-scale point cloud data sets.The introduction of ICP algorithm and NDT algorithm can complete the registration and splicing of multi-mode point cloud data.First,the point cloud data with low accuracy obtained by UAV and the point cloud data obtained by laser scanning were initially registered to obtain a better registration effect.Then,the point cloud data after NDT registration was further precisely registered by ICP algorithm,so as to establish the registration and fusion model of multi-mode three-dimensional point cloud.In summary,this thesis systematically studied the multi-mode three-dimensional point cloud data model generated,registration and fusion,is proposed based on the center of gravity and center of mass of the initial point cloud registration algorithm and verified the algorithm for large scale point cloud data sets,the effectiveness of the initial registration,and focused on the accuracy of point cloud data of UAV to obtain low question for related research.Finally,the registration and fusion of multi-mode point cloud data are completed by introducing NDT and ICP algorithm.