The Technology Of Terrain And Building Reconstruction Using Airborne Full-waveform LiDAR Data

Airborne full-waveform LiDAR is an emerging technology in the field of photogrammetry and remote sensing. Terrain and building reconstruction is one of its main applications. In this dissertation, the characteristics of full-waveform information and shortage of traditional methods are analyzed, and the waveform decomposition, point cloud filtering, building extraction and building reconstruction are researched based on existing research achievements. The main achievements and innovations are listed as follows:1. After elaborating the background and significance of terrain and building reconstruction, the systematic and technical principle of airborne full-waveform Li DAR are introduced. Furthermore, the specifications of some main systems are generalized. Finally, the current reseach situation and problems are concluded and analyzed.2. An iterative wavefrom decomposition method based on global convergent LM(levenberg marquardt) is proposed. In our method, the stratagy of iterative peak detection is adopted. During the waveform fitting process, the optimal estimation of Gaussian parameters is realized by using global convergent LM. The problems existing in traditional nonlinear least squares method are solved, such as local convergence and uncompleted peak detection. Experimental results prove that, the residual is smaller, the ratio of reliable pulse is higher and the level of point cloud is richer than that of traditional method.3. To begin with, the waveform information and robust estimation theory are introduced to detect abnormal points. Then, the terrain curve is fitted weightly according to waveform parameters. And the self-adaptive height difference threshold is set given the window size and mean square error. A weighted curve fitting filtering method fusing waveform information is proposed, which overcomes restriction of geometry feature and some drawbacks exising in traditional curve restrained filtering method such as seed selection and height difference threshold determination. Experimental results show that, the filtering error rate is lower and the reconstructed model is closer to real terrain by using our method.4. A building extraction method based on Latent Dirichlet Allocation(LDA) model is proposed. The point clusters are generated by super voxel segmentation to resolve the efficiency problem. The theme features of point clusters are obtained by LDA model to improve the accuracy. Experimental results prove that, our method not only increases the accuracy of building extraction, but also substantially reduces the consuming time, and it remains stable with different number of latent themes and vocabulary.5. Focus on the problem of building reconstruction, an automated approach for complex shape building reconstruction based on key point detection is proposed. By combining RANSAC segmentation and space segmentation, the point cloud of different roof planes are extracted automatically. For each plane, the exact contour is picked up using the Alpha Shape algorithm. And intersection line features are determined by topological relation of those planes, which will help correcting the initial key points. Finally, the precise building model is obtained. Experimental results show that our method could realize building model reconstruction of different structures. And the accuracy is comparable to TerraSolid’s results.