Accurately modeling of virtual plants is well cared in computer graphics. Plant modeling has an important significance in digital entertainment, eagroforestry, protection of Ancient and rare trees and digitized planning. In recent years, people have conducted in-depth researches in the field of plant modeling and gained certain achievement. Many reconstruction methods have been proposed. Until now, there is no efficient method that can solve this problem, since complex shapes and heavy occlusions are included, so that Faithful 3D reconstruction of living trees is still a challenge. We use point cloud of trees as input, and carry on research in modeling of virtual plants and segmentation of point cloud of trees. The main contributions of this thesis include following four issues： 1.We propose a method for the segmentation of point cloud based on the distribution of principal directions. Points from leaves and points from branches are separated based on a new metrics on the convergence of local directions. A region growing method is used to classify points from different branches, and region merge is adopted to avoid over-segmentation. Our metho needs no interaction during the process and brings less error segmentation. 2.We present a range image segmentation and twigs growth based approach to construct a 3D model of a real tree form a single range image. Point cloud are separated first. Then points from the same branch are further segmeneted along the axis direction of the branch. Main skeleton node is computed from each points group. The shape patterns of visible branches are used to predict those of obscured branches. The small branches generated in our method keep the biological features of main skeleton and perform the shape of crown. 3.We propose a new tree reconstruction technique faithful to their actual data. This method chooses single scanned point cloud as our data source and reconstructs 3d models base on shape analysis of 3D laser scan data. First principal directions and principal curvatures are estimated with quadric surface fitting. Then main skeleton nodes are extracted based on a space searching algorithm. Feature points of crown are extracted and hierarchical particle flows are constructed to organize feature points and to simulate twigs. This method is mainly automatic and easy to use, and it can be applied to model big trees with complex shapes and heavy occlusions. 4.We present an approach to reconstruct point cloud of a forest. Th...
修改评论