Forest Identification And Reconstruction Extraction From Airborne LiDAR Data

Airborne LiDAR is a new type of measuring system, which is being widely applied in the forest resources research and increasingly being regarded as an important means in obtaining and analyzing the model forest data. However, most of the existing methods target at the forest stand identification and parametric inversion. In order to satisfy the demand of more precise, automatic and meticulous, it is inevitable in modern forestry research to find out how to identify a single tree in high canopy density and such complex conditions. In the meantime, there is no automatic modeling algorithm of a single tree, which hinders wide spread of Airborne LiDAR in digital forestry. As a result, it is meaningful to devising an automatic and efficient system of identifying a single tree and modeling.Based on full waveform LiDAR data, with forest point cloud segmentation and three-dimensional modeling as its aim, obtaining precise and integrated forest point and cloud data as its focus, and the "full waveform LiDAR data processing-tree identification-forest point cloud segmentation-three-dimensional reconstruction of trees" as its main line, this paper conducts a research on the key technology in the following aspects:1. Establishing a set of technical framework based on LiDAR data to identify trees and reconstruct the tree model. First of all, making use of full waveform LiDAR data to obtain high-density point clouds; secondly, utilizing the image processing based on the crown elevation model to conduct a preliminary identification; thirdly, based on Markov Random Field and the Bayesian Theory to realize three-dimensional point cloud segmentation; then on this basis, to extract the trunk skeleton, and with a priori knowledge to generate a complete single tree skeleton; lastly, through the grid processing and texture mapping of the skeleton and leaves to reconstruct the tree model and draw the three-dimensional scenes.2. Proposing a means of full-waveform LiDAR data decomposition and obtaining high quality point cloud. The tradition method refers to the simple threshold provided by Equipment Manufacturers so that they can not obtain highly accurate data. This paper manages to solve such a problem. Firstly, based on the Generalized Gaussian function to decompose the radar waveform data; then, using Levenberg-Marquardt algorithm for parameter optimization in order to obtain a three-dimensional coordinate of high-density point clouds, at the same time, collect the amplitude of the point cloud, height and backscatter of cross-section and other important information which provide data support for the follow-up point cloud classification.3. Presenting a method of identifying point cloud and Extracting the crown edge of the tree. Firstly, extracting non-surface point cloud by linear foreseeing filter method, on this basis, classifying different forest point clouds according to point cloud distribution and waveform parameters; then, turning the forest point cloud into the CHM forest canopy elevation models, and marker-controlled watershed segmenation to locate the treetop and extract the crown, which provide a priori information for plant trees for subsequent precise segmentation.4. Proposing a three-dimensional point cloud segmentation algorithm. The traditional way locates s a single tree by using the canopy in the two-dimensional image, which works well in sparse woodland. However, in the high-density forest, plant trees often crowd together, which leads to low accuracy of traditional methods. In this paper, a new solution is proposed. Based on the anisotropic Markov model and three-dimensional point cloud segmentation method, by using Bayesian estimation theory to convert the problem of point cloud segmentation into the combinatorial optimization of posterior energy function; with Markov random fields, designing a priori anisotropy model based on the trees distribution and shape characteristics, and adopting Graph Cuts to optimize the combination of energy functions, then utilizing the approximate EM algorithm to iteratively calculate the model parameters.5. Proposing a method of reconstructing a three-dimensional tree model based on airborne LiDAR point cloud data. The traditional modeling approach requires considerable ecological knowledge, in addition, such approach is influenced by various parameters and it is difficult to control the model shape. This paper, combined geometric structure and biological morphology and on the basis of LiDAR point cloud, aims to reconstruct the tree model. First of all, utilizing the shortest path approximation algorithm to reestablish the in the connected graph with weight formed by scattered forest point cloud; based on trees’ self-similarity and information of leaves point cloud, combined with geometry and biological method, reconstructing a more complete and detailed skeleton of the trunk; using the cubic B-spline to build a more suitable trunk curve, and conduct the cylinder surface grid parameterization; lastly, in tangent space making use of the normal texture mapping and illumination model to obtain a single tree’s texture map so as to reconstruct a three-dimensional model.The experimental results show that the technical program has the following characteristics:(1) effectively improve the point cloud density, enhance the level of point cloud and obtain valuable spectrum eigenvalue of forest point cloud;(2) rapidly realize the monitoring of forest point cloud and locate a single tree;(3) effectively and efficiently achieve forest point cloud segmentation in high canopy density;(4) generate a relatively realistic tree model, with good randomness and controllability.In a word, this paper sheds new light on forest resource research and forest management, which makes a great contribution to the sustainable development of modern forestry.