【摘要】 为了实现预制拼装混凝土桥梁的虚拟预拼装，需要基于三维激光扫描点云对现场桥梁预制构件进行逆向建模，以将现场制作的实际构件转化为虚拟模型用于后续分析。针对施工环境下三维激光扫描点云的质量现状，重点对逆向建模结构的特征点定义以及特征点提取算法进行了研究，提出了一种针对预制混凝土梁构件几何特征的可适应于不同点云质量的自动逆向建模方法。该方法首先利用设计图纸转化先验数据对实测点云进行尺寸与位姿识别，并采用改进的迭代加权PCA方法精确计算实测点云中各点的特征法向量，并对各点进行所属结构面分类；然后生成建模特征点近邻区域点云的Voronoi图以分析点云分布的疏密度；最后针对不同结构面近邻区域点云的疏密度质量提出了不同的模型特征点提取方法，自动计算拟合坐标值，完成预制构件的逆向建模。采用该方法对单片预制混凝土组合箱梁的扫描点云进行自动逆向建模，计算建模特征点坐标拟合值，从中提取构件关键尺寸用以检验方法精度并与与现场实测值比较。结果表明：该方法可自动判断各结构面点云质量，能够根据不同点云质量选择对应的建模特征点提取原则计算拟合坐标值，所计算的几何尺寸最大绝对误差为6 mm,最大相对误差为0.4%,达到了较高的建模精度，具有一定的工程价值。更多还原

【Abstract】 In order to underlie the virtual assembly of precast concrete bridges, needed the reverse modeling based on LiDAR for the prefabricated components of on-site bridges, so as to convert the actual components made on site into virtual models for subsequent analysis. The quality of 3 D laser scanning point cloud obtained on site, concentrated on the definition of characteristic points and corresponding extraction algorithms of them, and an automatic reverse modeling method according to the geometric characteristics of precast concrete beam components was proposed, which could adapt to different quality of point cloud. In this method, the size and pose of the point cloud were firstly identified by transforming the prior data from the design drawings. Then the normal vectors of the points were accurately computed by the iterative weighted PCA, and each point in the measured point cloud was classified according to the normal vectors. The Voronoi diagrams of the point clouds in the adjacent areas of the modeling characteristic points were generated to analyze the distribution and density of these point clouds. Finally, extraction methods of the characteristic points for different density of the point clouds in different adjacent areas were designed, the fitting coordinates were computed, and the reverse modeling of the prefabricated components were completed. The reverse modeling method was used to automatically compute the fitting values for the coordinates of the modeling characteristic points, from which some key dimensions of the components were extracted to verify the accuracy of this method. The results show that this method can automatically analyze the quality of different structural panels, and correspondingly choose the appropriate method to compute the fitting values. The maximum absolute error of all the items is 6 mm while the maximum relative error is 0.4%. It has relatively high modeling accuracy for concrete beam components under actual engineering conditions and certain engineering value. 5 tabs, 14 figs, 23 refs.更多还原