Research On Extracting Three Dimensional Structure Information Of Urban Vegetation Using LiDAR Technique

The thesis is supported by Special Fund for Ministry of Land and Resources Research in the Public Interest (Research on Technology of Land Dynamic Monitoring,201011015-1), studying on the application of low altitude remote sensing technique. The typical vegetation with three-dimensional structure character was selected in downtown of Nanjing. Airborne LiDAR (Light Detecting and Ranging) and mobile laser scanning system point cloud data of the vegetation were used. The study based on LiDAR point cloud image representing of the vegetation structure, using the laser intensity based classification of vegetation, the three-dimensional spatial clustering of point clouds and changed threshold in search method algorithms, was for recognizing and extracting three-dimensional structure information of the urban vegetation.Urban vegetation as an important part of the evaluation of ecological city was concerned widely by more scholars. How to extract three-dimensional structural information of urban vegetation with remote sensing images that was a difficult problem to solve. LiDAR technique has capability of detection the three-dimensional structure of vegetation in vertical direction. It must be worthy of further study on the theories, methods and applications. The main contents and results of the research were presented as follows:1. The study presented image representations of three-dimensional structure of urban vegetation based on characteristics of vegetation structure and LiDAR technique. The LiDAR technique has a good capability of detection of urban vegetation in vertical direction. According to the analysis of the echo characteristics of LiDAR points,12.82%of the vegetation points in the study area showed penetrate ability of detecting vertical vegetation structure. LiDAR technology can detect low vegetation under tall trees, with a good space description of vegetation. The concept of "two layers, one strip and two boundaries" was put forward based on the spatial expression pattern of urban vegetation with LiDAR technique. Two layers are tree layer and shrub layer. One strip is layered strip between with tree layer and shrub layer. Two boundaries, one is Digital elevation model (DEM), and the other is Digital Surface Model (DSM). It has elaborated on LiDAR point cloud image representing of the vegetation structure. Using mobile laser scanner and field investigation methods, accuracy of two boundaries were evaluated as follows:1) RMS of the DEM with LiDAR was0.07m;2) Description of DSM by airborne LiDAR was0.28m lower than the actual materials. Some tree tops had more error, the maximum of which reached1.87m. The error of elevation value was small for artificial objects. Based on characteristics of urban vegetation grow and distribution, five three-dimensional structure parameters such as distribution of vegetation, canopy edge, crown shape, tree height and shrubs height were put forward. These contributed to further study on recognizing and extracting information.2. The study presented how to recognize three-dimensional structure information of urban vegetation based on image representations of three-dimensional structure of urban vegetation with LiDAR technique. The key step in identification of three-dimensional structure of urban vegetation was getting layered strip, for which classifying the LiDAR points was the first thing to do. A new classification method based on laser point intensity was developed, which could simplify complex mathematical calculations and distinguish between vegetation and non vegetation points effectively. The variation window and variation threshold algorithms were used to obtain three-dimensional layered strips between tall vegetation and low vegetation, and then identify the tree layer and shrub layer. Using field survey method, results of recognizing were evaluated as follows:the correct rate of the tree layer point was97.13%, and shrub layer point’s was77.63%. The local maximum window under a certain rule was used to search potential crown vertex so as to obtain the recognition result of tree height, and the height of tree-shaded shrub with the addition of information repairing interpolation method. Then, two types of vegetation, trees and shrubs were recognized.3. The study presented how to extract three-dimensional structure information of urban vegetation based on result of recognition three-dimensional structure information with LiDAR technique. The three-dimensional vegetation structure was divided into two aspects of planar structure information and vertical structure information. Vegetation outline, trees and shrub were firstly extracted. The space occupied by buildings, roads and water and other non-vegetation were called "vegetation empty". Extraction of vegetation empty was beneficial to obtain vegetation shape boundary. The roads were extracted based on the feature of point cloud distribution; the tree layer was extracted by the method of separating the point cloud followed by combining the point cloud, and using variation threshold algorithms, trees were searched and extracted based on tree crown outline instantiation with LiDAR point cloud and the assumption that the highest points were on behalf of the top of the tree; the shrub crown top surface was extracted as well by searching and interpolating. The result was evaluated with field measurement, and the RMS of shrub height with LiDAR was0.18m. Then the vertical structure information of vegetation was extracted, considering the planar information. The improved neighborhood interpolation algorithm of inverse distance power was used in extracting vegetation vertical top surface (also named normalized Digital Surface Model, nDSM). Tree heights were extracted by the algorithms of three-dimensional space cluster and the great value search of the variation window. The result was evaluated with field measurement and investigation. Tree height extracted by LiDAR was average0.27m lower, with the maximum of-0.89m. The error value of trunk position was average0.47m, the maximum of which reached3.04m. At last, three-dimensional shapes of tree canopy were fitted by the hybrid drive model. The result was evaluated with field sample investigation, and found that measurement of the shape of the canopy by LiDAR was basically accurate, but the diameter determination had errors.The research focused on spatial distribution of urban vegetation coverage by LiDAR remote sensing technology, applying new techniques to identify and extract three-dimensional structure of vegetation. The quantitative characteristics of three-dimensional structure of urban vegetation can be obtained with the effective methods established in the study when there was some shelter in vertical direction. The results of the research have an important impact on statistics and management of urban vegetation. The efficiency and accuracy of LiDAR technology should be further improved on its application of extraction of three-dimensional structure of urban vegetation, so as to put it into more extensive application.