Extraction And 3D Reconstruction Methods For Outdoor Ground Point And Indoor Structure Point Based On LiDAR Point Cloud Filtering

LiDAR point cloud 3D(three-dimensional)reconstruction is widely used in disaster prevention,urban planning and indoor navigation.However,the 3D automatic reconstruction of point clouds is quite difficult in scenes with complex indoor structure and outdoor terrain.In particular,the existing point cloud filtering algorithms need to set corresponding parameters according to different terrain features,and the automation of point cloud filtering is low.Moreover,the existing indoor room segmentation algorithms cannot handle special indoor structures such as slanting ceilings,and existing indoor structure model reconstruction algorithms cannot reconstruct interior structure details such as ceiling specific structure.How to adapt the 3D reconstruction algorithm to different structure features,improve the automation,and ensure the accuracy and completeness of the reconstructed model,is an important research problem in 3D reconstruction of point cloud.Starting from the point cloud 3D reconstruction process,this dissertation applies algorithms such as point cloud filtering to the extraction of outdoor ground points and indoor structure points,and the corresponding indoor and outdoor 3D models are reconstructed from the outdoor ground points and indoor structure points.The research content includes three aspects:outdoor ground point cloud filtering,indoor room segmentation,and indoor structure points extraction and 3D reconstruction.The research aims to further improve the reconstruction algorithm system of outdoor terrain model and indoor structure model.The specific research includes:(1)To address the problem that most of the existing point cloud filtering algorithms need to set parameters according to different terrain features to obtain the best filtering result,this dissertation proposes the point cloud filtering algorithm by the integration of multi-scale morphology and triangular irregular network.Firstly,the multi-scale morphological filtering algorithm is used to extract the initial ground points,and the improved radial nearest neighbor algorithm is used to perform local plane fitting to obtain more reliable potential ground points.Secondly,the potential ground points are used as ground seed points of the triangular irregular network filtering algorithm for ground point densification.Finally,continue to densify the irregular triangle network according to the height difference between the irregular triangle network and the remaining point cloud,and complete the classification of ground points and non-ground points.The proposed algorithm only needs to set the maximum operation window size of multi-scale morphological filtering algorithm according to the characteristics of the point cloud data,and the remaining parameters are fixed values.In the experiments of backpack LiDAR point cloud data and airborne LiDAR point cloud data,the proposed algorithm simplifies the algorithm parameter settings and improves the automation of point cloud filtering while ensuring the accuracy of point cloud filtering.(2)Aiming at the problem that the existing algorithms cannot handle special indoor structures such as slanting ceiling,this dissertation proposes an indoor room segmentation algorithm by integrating cloth simulation filtering and regular grid analysis.First,the algorithm uses the cloth simulation filter algorithm to extract the complete ceiling structure,and then use regular grid analysis to perform preliminary segmentation while removing possible outlier points.Secondly,the number of correct rooms is obtained using morphological erosion,and the adaptive threshold of neighborhood filtering is set based on the obtained number of rooms,which is used to deal with the under-segmentation caused by noise points between ceiling structures.Finally,repairing the boundary grid affected by neighborhood filter based on point cloud normal vector estimation to complete indoor room segmentation.Through experiments on indoor point cloud data acquired by different acquisition platforms,it is verified that the algorithm has high accuracy and strong robustness,and this algorithm deals with various indoor special structures.(3)To address the problem that most existing structure model reconstruction algorithms have poor accuracy and completeness when reconstructing special structures such as curved wall,this dissertation proposes an indoor structure model 3D reconstruction algorithm by fusion of model-driven and data-driven.First,all structure point clouds including ceiling,wall and floor are extracted by point cloud filtering and other algorithms.Among them,the ceiling boundary points are used as control points to extract the point cloud of the wall structure,which can better extract the special structure of curved wall.In addition,the slice-grid detection algorithm is proposed to extract the special structure of ceiling to add the structure details.Then,based on the different geometric features of the extracted structure point cloud,the corresponding optimization models are used to complete the model-driven structure point cloud refinement to reduce the impact of point cloud noise on the model accuracy.Finally,the model is reconstructed based on the data-driven reconstruction algorithm to retain all structure features.The experiments on reconstructing models from indoor point cloud data with different structure features show that the algorithm can improve the accuracy and completeness of reconstructing structure details such as curved walls,and the real situation of the indoor environment can be better described.There are 76 figures,17 tables,and 179 references in this dissertation.