Three Dimensional Geometrical Modeling Of Individual Tree From Terrestrial LiDAR Scanner Point Cloud

Three dimensional (3D) geometric modeling of individual tree is an important and challenging research topic in the field of virtual plant. Laser scanning technology provides a new tool that can rapidly and accurately acquire the 3D coordinates of plant (trees)surface features, which provides a significant source of data for realistic and high-precision tree modeling. Based on the principle of geometrical topology and tree architecture, this thesis presents an efficient 3D geometric tree model reconstruction approach that is able to create precise and photorealistic individual tree model using point cloud data obtained from terrestrial laser scanner (TLS). We focused the research on some key techniques such as skeleton extraction, classification of tree branches, enhancement of model realism, interactive editing, and the extraction of tree growth and geometric parameters; then we integrated these tools and built an application software system. The specific contents and results are as follows:(1) Based on the analysis of the drawbacks of traditional 3D plants (trees) modeling methods, and the merits of 3D trees reconstruction from point cloud data, we propose the research idea and the technical route of 3D reconstruction with high geometry accuracy and interactive editing. The existing algorithms for skeleton extraction from point cloud were analyzed and compared.(2) We propose ITSEPC, an innovative way of extracting individual tree skeleton from point cloud data. Skeletons of both thick and small branches can be extracted from the point cloud in a single run. Depending on the different circumstance of whether the branches are obscured by leaves and other tree parts or not, ITSEPC can deal with both complete and incomplete data. For the main branches that are not easily blocked and the data is more complete, skeleton of branches are extracted directly; but for those twigs which are too small to be scanned and are frequently obscured by other leaves, data acquired is often missing and incomplete. Because the point cloud information of leaves reflects the overall shape of the canopy, we treat the leaf cluster center as the skeleton point of the twigs, thus inferring the spatial distribution information of those twigs. This modeling approach has some advantages that it is of high efficiency, a strong sense of reality, close approximation with the morphological structure of real tree, and it is convenient to be integrated seamlessly with the mainstream parametric individual tree modeling software.(3) We propose a method to automatically classify and group tree branches’ corresponding skeletons into different levels, and realize 3D geometric reconstruction in a hierarchical form. Combined with the calculated radius information of the skeleton, we conduct 3D model fitting for the identified hierarchical skeleton using generalized cylinder. Furthermore, in order to enhance the fidelity of the tree model, we introduce the OpenGL rendering and bump texture mapping techniques. In addition, we conduct qualitative assessment and quantitative evaluation for the generated 3D tree model. Our experiments show that high precision is achieved for main branches with little view obstruction (the average error is 18.4 mm); however in branching regions, the model precision is relatively low. In crown area, compared with the bounding contour of real point cloud, the modeled tree canopy matches up to more than 90% of area.(4) We conduct the studies on interactive editing of 3D geometrical tree model and extraction of morphological parameters from point cloud. Based on the data structure of hierarchical tree topology, the techniques of OpenGL double buffering and 32-bit RGBA color coding, we realize the interactive branch picking function so that a user can select a branch quickly and accurately, and the user can further adjust the branch radius or delete branch. By changing the parameters of some organs such as leaf size, leaf density, and leaf number, we make tree model resemble more to the real tree. In addition, we propose a method for tree morphological parameters extraction. By comparing these extracted parameters with the measured field data, we have verified the effectiveness of our methods.(5) Based on the parametric individual tree modeling tools ParaTree, we have designed and implemented a prototype 3D geometric modeling system PC2Tree (Point Cloud to Tree) by integrating all the techniques mentioned above. The system is able to not only reconstruct the 3D model with high precision, but also supports interactive editing of tree model and rapid extraction of morphological parameters. The fully functional system has user-friendly interface and provides a useful tool for the applications such as tree growth simulation, dynamic monitoring and resources evaluation for forest, famous and old tree preservation and management. Finally, taking a camphor tree in Fuzhou city as an example, the whole procedure from data input to the model generation and output is described to demonstrate the system function and performance.