Tree Point Cloud Data Processing And 3D Modeling Based On Terrestrial LiDAR

With the rapid development of computer vision technology and the constant update of Terrestrial Laser Scanning(TLS),TLS has been widely used in building surveying,heritage conservation,medical treatment,forest resource survey and other fields.In forest resource survey,TLS is a non-contact and non-destructive 3d measurement technology,which can generate point clouds with different resolutions from multiple perspectives and has the potential to automate data processing.It has been widely used in the extraction of forest structure parameters and the construction of digital forest models.It is of great significance for the accurate improvement of forest quality and the level of China's forest resource information management.In this study,the fast point feature histogram algorithm and the normal distribution transformation algorithm were used to realize multi-station forest land point cloud registration.Then,the octree algorithm was used to compress the point cloud data of the forest land,and the random sample consensus algorithm was used to complete the segmentation of ground point and tree point.On this basis,Euclidean clustering and the minimum cut algorithm were used to segment the tree point cloud,and the structural parameters were extracted from the segmented tree.Then,super-voxel clustering algorithm was used to segment the tree branches and leaves.Finally,three-dimensional reconstruction of individual tree branches and leaves point cloud data was carried out.(1)In order to improve the efficiency and accuracy of 3D point cloud data registration in massive forest land,this study proposed a registration algorithm based on the combination of fast point feature histogram and normal distribution transformation.The experimental results showed that the average registration error of the algorithm(the average distance of the corresponding point pairs)was 0.0244 m,and the running time was 243.512 s.Compared with the traditional ICP algorithm,the algorithm had greatly improved the registration accuracy.It can well achieve cloud registration of multiple measurement sites in forest land.(2)Euclidean clustering algorithm and minimum cut algorithm were used respectively to segment the TLS point cloud data of forest land,and the integrity rate of individual tree in the two algorithms was also compared.The parameters of individual tree structure based on two segmentation algorithms were extracted and analyzed by regression with the measured values.The experimental results showed that both Euclidean clustering algorithm and minimum cut algorithm realized the segmentation of individual tree point cloud data of sample plot,but the minimum cut algorithm was better than Euclidean clustering algorithm.The structural parameters of the individual tree based on the minimum cut algorithm were highly correlated with the actual measured values,and the determination coefficients of DBH,tree height and crown width of individual tree were 0.969,0.810 and 0.763,respectively.(3)A super-voxel segmentation and greedy projection triangulation based on kd-tree texture mapping algorithm were proposed for combining tree point cloud reconstruction methods in this study.The experimental results showed that the algorithms could realize the construction of the detailed of the branch surface and leave models,and restore the true tree shape structure and texture color.By testing the tree point cloud data of different densities,it was found that the main branch structure of the tree model was well preserved,which indicated that the modeling algorithm is robust to the change of point density.The improved algorithms in this study realize the registration of point cloud scanned from forest land,the segmentation of individual tree and the three-dimensional reconstruction of individual tree.These research results have strengthened the application of terrestrial laser scanning technology in the dynamic survey of forest resources,and provide effective reference for data processing of cloud registration,single tree segmentation and modeling in forest land.