Visual Simulation Study Of Single Wood Based On Terrestrial LiDAR

The technology of visual representation of the morphological structure of single trees has important theoretical significance and application value for precise forestry management.The segmentation as well as visualization of single wood remains a challenging topic in the natural environment where trees grow in complex environments with overlapping branches and leaves.Ground-based Li DAR is used to acquire point cloud data from two plantation forests and a campus street tree in this thesis.The single tree point cloud was first segmented,and then the skeleton of the tree was extracted,so as to build a 3D model.The main research elements of this thesis are as follows.(1)The collection and pre-processing scheme of the point cloud data of plantation forests was designed to address the problem of complex forest environment and the difficulty of resource survey.The point cloud data were collected on the campus of Nanjing Forestry University and the Baima forest farm,respectively.The Trimble X7 was fixed to a ground triangle and multiple stations were distributed within the sample plots to obtain the complete plot point cloud data.To address the problem of a large number of ground points in the fused plot point cloud,the RANSAC algorithm were used to remove the ground plane.Noise filtering was performed on the point cloud based on PCA dimensionality reduction.And the segmented ground points were used to generate a ground point grid to normalize the point cloud.The pre-processing of the point cloud data is the basis for the subsequent extraction of single wood and the construction of the3D model.(2)After the ground points were removed,the single wood was segmented using the spatial distribution rules of the point cloud tree and the overall segmentation accuracy of the two broad-leaved plantations reached 0.819 and 0.845,and the overall segmentation accuracy of the street trees reached 0.952.Subsequently,tree height was calculated using the height difference of the tree Z-values,the method of extracting the cylindrical model was applied to tree trunk extraction,and the point cloud slice was fitted with a circle based on least squares to calculate tree diameter at breast height.Considering that the trunk and leaves of broad-leaved trees are distributed in a faceted manner,when using point cloud geometric features for branch separation,the tree point cloud was first divided into a coarse branch collection containing large leaves,and a fine branch collection containing small leaves,finally Euclidean clustering was performed on them respectively.Experiments showed that the separation idea in this thesis could effectively extract the wood part of cherry trees.(3)After obtaining the single wood point cloud,the spatial competition strategy was used to extract the skeleton point of the tree for subsequent modelling.The working mechanism of the original space colonization method is to construct the branch architecture using simulated points randomly and uniformly distributed in space.In contrast,the point cloud tree,as the input information of the algorithm,constrains the spatial distribution of points at the outset.To address this problem,a DBSCAN-based space colonization method is proposed to obtain the skeleton of the tree,which acts as a pre-calculation of the number of branches before generating new skeleton points.After the introduction of DBSCAN algorithm,the improved extracted skeleton was evaluated in terms of branch angle,trunk length and primary branch length,with R~2 of 0.952,0.981 and 0.984,respectively.The RMSEs were 3.309degree,0.069m and 0.051m,respectively.The experimental results showed that the improved method extracted a skeleton with a better match to the scanned point cloud and a clearer topology.(4)After obtaining the point cloud tree skeleton,the wood structure of the tree was constructed using generalized cylinders and the leaf model was reconstructed using NURBS surfaces.Under Visual Studio platform,the model was rendered using Open GL,and the MFC framework was used to implement a visual interactive interface,tree models will display on the window.