融合地面与机载LiDAR的建筑物三维重建

发布时间：2018-06-17 09:37

本文选题：多源LiDAR数据 + 平面拟合　；参考：《西南交通大学》2014年硕士论文

【摘要】：建筑物信息对城市的建设和规划,社会的可持续发展有着重大影响,如何快速、有效地获取城市建筑物三维空间信息成为目前研究的热点。本文针对数字城市建设建筑物三维建模中如何构建各个立面的问题,融合地面与机载LiDAR点云数据,采用全局阈值平面拟合滤波算法,精确地提取出建筑物平面目标,根据空间几何信息对建筑物进行质量检测与几何校准,实现建筑物的三维模型构建。具体研究工作如下：(1)研究基于特征线的半自动点云数据配准融合方法。重点研究了半自动配准融合方法中的特征线定位,对方法进行改进,利用法向量约束求取平坦度得到平坦度差值,提取特征线。结合实验数据的特殊性,在地理概略粗定位的基础上,利用ICP算法拼接特征线,实现半自动数据配准融合,通过实验验证算法的可行性,对不同特征线下的配准效果进行分析比较。利用地理概略粗定位与特征线精定位的方法能较好地进行地面与机载LiDAR点云数据配准融合,配准融合的效果与选取的特征线密切相关。(2)改进一种利用斜率进行平面拟合的滤波算法。在归纳总结国内外学者在相关问题的研究基础之上,对该算法的阈值处理部分进行改进,针对算法中阈值具有模糊性等问题,预先对建筑物平面分类,利用总体最小二乘法的思想进行全局阂值估计。通过实验分析比较算法改进前后的精确度,分析其可行性。采用总体最小二乘法的全局阈值平面拟合滤波算法可以优化平面拟合的过程,并能一定程度地提高平面拟合的准确性。(3)实现实验数据的三维重建。选择具有代表性的实验数据,通过对建筑物平面的空间几何关系进行分析,对平面的平整度与倾斜度进行计算,在此基础之上实现平面的空间校准,对校准后的建筑物平面进行三维重建。进行空间几何关系校准后的建筑物平面具有精度更高、适应性更强的特点,更好地反映了建筑物的细部空间三维信息；融合地面与机载LiDAR点云数据的建筑物三维重建能更精细地反映建筑物真实三维信息,但其自动化程度仍有待提高。研究表明,基于特征线的半自动点云数据配准方法能较好地融合地面与机载点云数据,但该方法明显受所选取特征线的影响；改进的总体最小二乘平面拟合滤波法可以提高平面拟合的准确性；结合地面与机载点云的建筑物轮廓信息提取能较真实的反映建筑物的三维信息。
[Abstract]:Building information has great influence on the construction and planning of cities and the sustainable development of society. How to obtain 3D spatial information of urban buildings quickly and effectively has become a hot research topic at present. In this paper, aiming at the problem of how to construct each facade in the 3D modeling of digital city building, the point cloud data of ground and airborne LiDAR are fused, and the global threshold plane fitting filtering algorithm is used to extract the building plane target accurately. According to the spatial geometric information, the quality of the building is checked and calibrated, and the 3D model of the building is constructed. The main work is as follows: (1) A semi-automatic point cloud data registration and fusion method based on feature lines is studied. The feature line location in semi-automatic registration fusion method is studied, and the method is improved. The flatness difference is obtained by using normal vector constraints, and the feature line is extracted. Combined with the particularity of the experimental data, on the basis of rough geographical location, the ICP algorithm is used to join the feature lines to realize the semi-automatic data registration fusion, and the feasibility of the algorithm is verified by experiments. The registration effect under different characteristic lines is analyzed and compared. The method of rough geographical location and fine feature line location can be used for the registration and fusion of ground and airborne LiDAR point cloud data. The effect of registration fusion is closely related to the selected feature lines. On the basis of summing up the research on the related problems, this paper improves the threshold processing part of the algorithm, aiming at the fuzzy threshold in the algorithm, classifies the building plane in advance. The global threshold value is estimated by using the idea of total least square method. The accuracy and feasibility of the improved algorithm are analyzed and compared by experiments. The process of plane fitting can be optimized by using the global threshold plane fitting filtering algorithm based on the global least square method, and the accuracy of plane fitting can be improved to a certain extent, and the 3D reconstruction of experimental data can be realized. Select the representative experimental data, through the analysis of the spatial geometric relationship of the building plane, calculate the flatness and inclination of the plane, and realize the spatial calibration of the plane on this basis. The three-dimensional reconstruction of the calibrated building plane is carried out. After the calibration of the spatial geometric relations, the building plane has the characteristics of higher accuracy and stronger adaptability, which better reflects the three-dimensional spatial information of the building. The 3D reconstruction of buildings with ground and airborne LiDAR point cloud data can more accurately reflect the real 3D information of buildings, but the degree of automation still needs to be improved. The research shows that the semi-automatic point cloud registration method based on feature line can fuse the ground and airborne point cloud data well, but this method is obviously affected by the selected feature lines. The improved least square plane fitting filtering method can improve the accuracy of plane fitting, and extract the building contour information from ground and airborne point clouds to reflect the 3D information of buildings.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TU198

【参考文献】