Research On Airborne LiDAR Full-Waveform Data Decompositon And Point Cloud Classification

Airborne LiDAR,as an active remote sensing technique,can directly measure the range between the laser scanner and the Earth's topography,providing 3d point coordinate information.LiDAR data are known to be useful in many specific applications such as forest parameter estimation,3D city modelling and power line detection or Digital Terrain Model generation.Since 2004,new commercial LiDAR system called full-waveform LiDAR have appeared with the ability to record the complete waveform of the backscattered signal echo.The echo waveform is the result of interaction of the emission pulse with the target,it reflects the vertical distribution,geometric and physical characteristics of the target.Further physical properties of object may be derived with an analysis of the backscattered waveforms.It is very of great significance to further study the application of waveform parameters in vegetation extraction and classification of land cover.In this paper,we use the waveform parameters derived by the decomposition of the waveform to study the advantage of waveform parameters in point cloud filtering and feature classification application.The main researches of this thesis are as follows:1.The decomposition of waveform.Waveform Gaussian decomposition is the mainstream method for the full-waveform data process.we proposed a lateral Gaussian decomposition method,which can effectively estimate multiple overlapped waveform.after removing invalid components,we use Levenberg-Marquardt method further optimize the parameters.Experiments show that this new method can effectively detect different kinds of complicated waveform and can produce more high quality point cloud with additional waveform parameters.2.Point cloud filtering using waveform feature.Different objects have different waveform feature.We choose the wave width as a weight of point to improve the accuracy of filtering.First,we use the width threshold to remove the obvious non-ground points,and then use the width information to calculate the weight of each point,we realized a hierarchical weighted surface filtering method to create high precision terrain reconstruction.Experiments show that this method can effectively remove non-ground points such as low vegetation,and improve the correct rate of filtering.3.Point cloud classification using waveform features and geometric features.On thebasis of the waveform decomposition,the decision tree is constructed by training the selected sample data using the obtained parameter information(intensity,width,number of echoes),combining curvature and elevation difference geometric information.We classified the non-ground points to three land cover types,building,tree and low vegetation.Experiments show that the waveform feature can help to improve the extraction precision of vegetation.