Tree Skeleton Extraction And Modeling Based On TLS Data And Their Applications

Building a high-precision and high-fidelity tree 3D model can provide technical support for extracting tree height,diameter at breast height,branch attributes and other parameters,lay the foundation for estimating forest stock,forest carbon storage and other applications,and is the foundation and key for the in-depth development of forestry informatization.This article takes six tree species studied in the public dataset and three strains of American black poplar scanned in the field as the research objects.Using their Terrestrial Li DAR Scanning(TLS)data(i.e.individual tree point clouds),a kd-tree based single tree point cloud branch and leaf separation method is used to obtain point clouds for each branch;In this paper,a tree skeleton extraction algorithm based on incomplete simulation of water and nutrient transport in trees,ISTTWN algorithm,is proposed,and a new tree skeleton optimization scheme is constructed;Research has improved the method for estimating the radius of each branch of a tree,and based on this,a three-dimensional model of the tree has been constructed.The transformation from a discrete tree point cloud model without topology to a continuous generalized cylindrical model with topology has been achieved,presenting the three-dimensional model of the tree in a visual form.Combined with part of the data set collected,the tree skeleton and model are respectively applied to the extraction of tree branch attributes and single tree measurement factors.The main research content and achievements are as follows:(1)Acquisition of experimental data.This article uses both public and collected datasets.The public dataset is sourced from resources publicly available on the internet in existing research,including six tree species:Beech(Fagus sylvatica L.),Ash(Fraxinus excelsior),Oak(Quercus robur L.),Douglas Fir(Pseudotsuga menziesii(Mirbel)Franco),Spruce(Picea abies(L.)Karst.),and Pine(Pinus sylvestris L.).The collected dataset consists of a total of 36 sample plots of American black poplar(Populus deltoides Marshall),including Populus 95,Populus 895,and Populus 797,scanned by RIEGL VZ-400i in winter,a season when trees have few leaves.The sampled point cloud data was preprocessed through multi-station stitching,point cloud denoising,ground normalization,and single tree segmentation to obtain various wood point clouds.(2)Acquisition of branch point clouds.This article is based on the kd-tree structure of 3D point cloud data to remove duplicate points in a single wood point cloud.Based on the characteristics of Li DAR technology and the proximity zone of points in a single wood point cloud,the effects of branch and leaf separation under four commonly used radius parameters are compared.The experimental results show that the use of kd-tree has high efficiency in removing duplicate points.The branch and leaf separation of single tree point clouds based on kd-tree is generally better when the radius parameter is 10 times the average density of point clouds,and a relatively complete branch and leaf point cloud can be obtained.(3)Extraction of tree skeleton.In this paper,a tree skeleton extraction algorithm based on incomplete simulation of water and nutrient transport in trees(ISTTWN)is proposed according to an ecological research conclusion.The ISTTWN algorithm proposed relies on the information of the shortest path distance to distinguish the explored points and unexplored points in the branch point cloud.The branch where the unexplored points are located is distinguished by using the European clustering based on kd-tree,and the tree skeleton is constructed by recursion.In order to improve the efficiency of tree skeleton extraction,the ISTTWN algorithm is improved by using iterative acceleration and multithreading,and two post-processing methods,breakpoint connection and ring breaking,are combined.The experimental results show that the improved ISTTWN algorithm is implemented when the minimum fixed search radius(an infinitesimal amount is required in actual programming)that can make every point in the graph have neighbors is selected as the step size,and the initial skeleton of all kinds of trees is the most refined,ensuring the integrity,continuity and correct topology of the tree skeleton,but there are four shortcomings:The skeleton at the stump is twisted;The connection at the bifurcation is stiff;Too many unnecessary branches are existed;And the direction of the branches is abrupt.Therefore,further optimization is needed.(4)Optimization of tree skeleton.In this paper,aiming at the shortcomings of the initial skeleton,a tree skeleton optimization scheme is proposed,including the proposed method,skeleton reconstruction at the stump,that uses height information to replace the original partial skeleton by local reconstruction of the skeleton at the stump;the bifurcation optimization method based on cosine similarity and local point cloud information,and the pruning and existing skeleton smoothing methods.The experimental results show that after optimization,the skeleton at the stump is no longer twisted,the skeleton at the bifurcation and branches is natural,and unnecessary branches have been removed.The optimized tree skeleton is more consistent with the real topology of trees.(5)Establishment of tree models.On the basis of the existing optimized tree skeleton,a new method for estimating the radius of each segment of tree branches and trunks is studied and improved,and a new method for building a three-dimensional tree model is built on this basis.In this paper,considering the phenomenon of overfitting of the radius calculated by the least squares fitting circle at the bifurcation and the large bending of branches,radius derivation is adopted,and based on an accurate position of the radius calculation,the WBE model is used to estimate the radius of a single branch combined with energy stability and mechanical support stability,and its inverse solution formula.Based on Murray et al.’s proven plant growth formula for estimating the multi branch radius and Shinozaki et al.’s proposed pipeline model,the derivation of radii towards tree tops and root necks is achieved.When establishing the model,an arterial snake like model based on a three-dimensional convex hull was proposed without the need of node-to-node correspondence.The experimental results show that compared with the tree model established by the open-source AdQSM,the tree model obtained by this method is more in line with the input branch point cloud shape,and the output model file is smaller.(6)Application of tree skeleton and model.This article randomly selects a portion of collected poplar tree samples and uses measured values to test the performance of research results in extracting branch attributes and individual tree measurement factors.Specifically,in terms of tree factor measurement for individual trees,the open source AdQSM was used for comparison.The experimental results show that in terms of branch attribute extraction,the R~2 of the chord length and branch length extracted by this method is about 0.985 and 0.990,achieving good results,which are very close to the measured values.The R~2 of the extracted landing angle reaches a usable level at 0.75;In terms of single tree tree measurement factors,compared with the model established by the AdQSM,the model established in this article is more in line with the shape of the branch point cloud.The R~2 of the extracted diameter at breast height and volume reached 0.962 and 0.947,respectively,and the fitting effect of the model is better than that of AdQSM’s extraction results.The R~2 of the volume extracted by this method is superior to the R~2 of the existing poplar volume calculation methods in the research area,and is close to the R~2 of various fitted binary volume equations,with significant improvement.