The Technology Research In Reconstruction Of Three-dimensional Model Of Individual Broad-leaved Tree Based On Multi-source Optical Data

Three-dimensional modeling of vegetation is currently one of the hot issues in the research field of resources and environment. The emergence and development of three-dimensional laser scanning technology provide a new data acquisition approach for the establishment of three-dimensional model of vegetation. Based on the acquisition of three-dimensional laser point cloud data and hyperspectral data, this paper studies the methods and approaches for the three-dimensional model reconstruction of an individual magnolia tree using multi-source optical data. The main research works are as follows:(1) This paper establishes a relatively complete process of model reconstruction for an individual magnolia tree. Firstly, based on the relevant theories of three-dimensional laser scanning technology and hyperspectral imaging technology, the acquisition methods of multi-source data and the operation regulations of the instruments are introduced. Then the obtained three-dimensional point cloud data of the magnolia tree is orderly organized and indexed, filtered and denoised, segmented for branches and leaves, and multi-station registered. Next, the trunk and branches of the magnolia tree are respectively modeled by different methods because of their distribution difference of point cloud density. And by adding spectral information in the geometric modeling process, a diversified leaf model with geometric and spectral information is established. Eventually, the three-dimensional model of the magnolia tree which can reflect the actual geometric structure parameters and the leaf spectral characteristics is established.(2) Considering the feature of large point cloud density of the magnolia trunk, this paper firstly extracts the convex hull line from the isometrically layered trunk point cloud using convex hull algorithm, then builds the triangular meshes between adjacent layers and finally obtains the three-dimensional model of the magnolia trunk by merging the triangular meshes between adjacent layers. This method could effectively compress the quantity of point cloud, and improve the processing efficiency of point cloud data. The constructed magnolia trunk model could reflect the detailed geometric characteristics of the trunk surface, but the parameters of circumference of the truck model are larger compared with the actual measured data. The main reason for this error is analyzed and the suggestions for accuracy improvement are given.(3) Considering the small point cloud density of the magnolia branches, large data error or lack of some branch point cloud data, this paper firstly obtains the parameters of center and radius of the branch skeleton in each layer using the least squares method, then links the skeleton center of each layer to obtain the branch skeleton, and finally fits the curved surfaces of branches according to the radius of each skeleton layer. The constructed branch model could reflect the real space topology of the magnolia branches, but the length parameters of branch model are larger compared with the actual measured data. The main reason for this error is analyzed and the suggestions for accuracy improvement are given.(4) This paper introduces the hyperspectral data of the magnolia leaf as auxiliary modeling data. In the process of geometric modeling of the magnolia leaf using point cloud data, the hyperspectral data and the point cloud data of the selected magnolia leaf are registered to eventually obtain the geometric-spectral leaf model that contains multi-source information. The research and application field of three-dimensional model of plant could be further expanded in this way.