基于密集点云重投影的三维线段模型重建算法  

3D line segments reconstruction algorithm based on dense point cloud re-projection

作　　者：欧元汉[1] 黄鸿[1] 陆宇海 

机构地区：[1]广西大学计算机与电子信息学院,广西南宁530004 [2]南宁有无科技有限公司,广西南宁530007

出　　处：《计算机工程与设计》2015年第12期3291-3296,共6页Computer Engineering and Design

摘　　要：为实现大规模建筑物三维线段模型重建,提出一种基于密集点云重投影的三维线段模型重建算法。由运动恢复结构算法和多视图三维重建算法得到密集点云;在二维线段匹配阶段,将密集点云重投影到各视图中,根据关联同一空间点的二维线段存在弱匹配关系原理构建匹配线段集;在三维线段生成阶段,使用反向投影平面和反向投影射线相交法确定三维线段端点,使用RANSAC算法去除错误线段,以最小化三维线段重投影误差方法优化场景模型。实验结果表明,该算法的重建结果精确且具有良好的视觉效果。A dense point cloud re-projection based 3D line segment reconstruction algorithm was proposed to extract 3D line seg- ment model of large-scale urban building scenes. Dense point cloud was generated using structure from motion algorithm and multi-view stereo reconstruction algorithm. In line segments matching stage, dense point cloud was re-projected to multiple views and line segments that related to the same 3D point were matched. In 3D line segments generation stage, the end points of 3D line segments were obtained using intersection of back-projection plane and back-projection line, RANSAC was used to elimi- nate ill line segments. The whole scene model was optimized by minimizing re-projection errors of line segments. Experimental reults show that the algorithm produces accurate reconstruction result with satisfied visual effects.

关 键 词：三维线段 多视图 重投影 线段匹配 RANSAC 

分 类 号：TN911.73[电子电信—通信与信息系统]