The Research On Single Broad-leaved Tree Visualization Simulation Based On Three-dimensional Point Cloud

The accurate three-dimensional reconstruction of a tree model is of great significance for its growth simulation.With the rapid development of the virtual technology,tree 3D visualization based on the measured tree data has become a hot topic in the virtual plant research.This paper took the 3D point cloud data of single-broad trees collected by Terrestrial Laser Scanning(TLS)as objects,and several different algorithms were proposed,which are: the leaf and wood separation algorithm combining LCCP and K-means++ clustering;the filtering method for leaves based on manifold distance and normal estimation;the tree trunk and branch skeleton extraction and growth based on the horizontal stratification and normal-constrained space colonization algorithm.Based on these algorithms,the morphological structure simulation and model reconstruction of a single broad-leaved tree were primarily realized.Moreover,this paper used OpenGL to improve the authenticity of the model.The NURBS surface was used to fit the surface,and the generalized cylinder was used to draw the tree trunk and branch.The single broad-leaved tree visualization system was also developed to show the whole reconstruction process of different tree species.The specific researches and results are as follows:(1)The collection and preprocessing of the broad-leaved tree point cloud data.This paper used Trimble TX8 to collect point cloud data of different trees in different sizes and shapes.After scanning,the point cloud data was preprocessed.First,this paper analyzed the advantages of the manifold space and the manifold distances between sample points were calculated.Then,based on these distances,the target single tree was separated from the complex background using the pass-through filtering and denoised quickly using Gaussian filtering separately.This paper analyzed the filtering effect with different numbers of points in a neighborhood in Gaussian filtering and selected the optimal value.The point cloud after preprocessed can retain the real morphological structure and characteristics of the original broad-leaved tree to a great extent,and filter out most noise,which is conducive to the subsequent processes.(2)The leaf and wood separation algorithm combining the LCCP algorithm and K-means++clustering.First,the LCCP algorithm was used to realize over-segmentation.This paper took the characteristics of the broad-leaved tree point cloud into great consideration and creatively constructed the 37-dimensional feature space,which introduces the curvature difference into the similarity distance calculation and combines it with positions of points and FPFH,to judge the similarity between each voxel and complete the over-segmentation of the broad-leaved tree point cloud.Then,the supervoxels obtained after over-segmentation were further clustered based on the CC criterion and SC criterion.Second,the K-means++ algorithm was used to segment point cloud of leaves and branched which were still overlapped and connected after the LCCP algorithm.The result shows that this method can achieve high robustness and effectiveness in dealing with broad-leaved trees in different sizes and densities.(3)The filtering method for single broad-leaved tree leaves based on the normal estimation.This paper summarized the shortcomings in existing filtering methods,and redefined points need to be removed,which were outlier clusters,outlier points and noise points.Outlier clusters are those whose distance between cluster center and other cluster centers is greater than the threshold;Outlier points are the points far from other points in the same cluster;noise is the points whose normal directions are different from others.The normal direction was calculated based on the manifold distance calculation and the principal component analysis.The adaptive truncation was used to calculate three thresholds.The experiment results show that leaves in different forms are both intact and can keep original shapes after filtering.(4)The tree trunk and branch skeleton extraction and growth based on the horizontal stratification and normal-constrained space colonization algorithm.This paper analyzed the point cloud features of several broad-leaved trees and found that the point cloud of trunks is different from that of branches.The distribution of the point cloud of a trunk is uniform,while the point cloud of a branch is different both in density and volume.Hence,different extraction methods should be used in different parts.For trunks,point cloud was projected to a 2D plane in the middle height after horizontal stratification,and points on the plane were fitted as a circle to obtain the skeleton and the radius.The method is fast and accurate.Besides,the paper discussed the influence of different layer thicknesses in horizontal stratification.For branches,a new algorithm based on the normal-constrained space colonization was proposed.This algorithm took the skeleton of the layer where the branch grows as the input.To realize the automatic recognition and growth of different branches,the normal direction was applied in growth vector calculation to provide the traction force.To improve the authenticity of the skeleton growth,the paper also analyzed the automatic skeleton growth effects under different search radii and proportions of the normal vector in the growth vector.(5)The 3D reconstruction of the single broad-leaved tree based on NURBS surface and generalized cylinder.This paper combined C++ with OpenGL to realize the 3D reconstruction and visualization of a single broad-leaved tree.For leaves,36 control points were selected from the leaf point cloud after filtering,and the NURBS surface was used to realize 3D structure reconstructions.Then,a generalized cylinder was generated by drawing and rotating branch elements between every two skeletons to describe the 3D structures of trunks and branches.This paper also designed the visualization system interface to show the whole reconstruction processes of different single broad-leaved trees.