基于LiDAR点云数据的3D道路曲面快速重建方法研究

3D road surface is the core geographic element for forming the virtual landscape of digital city. There are still some issues in the study of reconstructing 3D road surface based on LiDAR point cloud, such as incomplete noise filtering, low computational efficiency, insufficient surface smoothness. Aiming at the above issues, this project mainly focuses on the following research contents by constructing the dimension-reduced mapping, introducing the robust regression models and spline smoothing methods: (1) construction of the dimension-reduced mapping for transforming the 3D LiDAR point cloud into the two-dimensional plane; (2) improving robust regression models for denoising the 2D point cloud and producing the fitted 3D road boundary curves; (3) developing innovative approach of picking out the triangulation nodes adapting to the road surface curvature variation and fast reconstructing 3D road surface via the spline smoothing methods. Based on 3D LiDAR point cloud, this project intends to systematically establish an all-in-one approach integrated with the core functions of “point cloud filtering”, “road boundary fitting” and “surface smoothing” by fully combining with the road surface characteristics, introducing the robust regression models and spline smoothing methods, thereby fundamentally improving the speed and quality of 3D road surface reconstruction. Research outcomes contribute to promote the further development of research on LiDAR point cloud data processing and 3D object reconstruction.

3D道路曲面是构建数字城市虚拟景观的核心地学要素。基于LiDAR点云的道路曲面重建研究目前尚存在点云噪声滤波不彻底，算法效率过低，重建曲面不够光滑等问题。本项目针对上述问题，通过构造降维映射、引入鲁棒回归模型与样条平滑方法，重点研究以下内容：（1）构造将3D LiDAR点云变换到二维平面内的降维映射；（2）建立适合处理降维后2D LiDAR点云的鲁棒回归模型，完成点云去噪，并拟合出3D道路边界曲线；（3）构建自适应道路面元曲率大小变化的三角剖分格网以及利用样条平滑方法快速重建3D道路曲面的算法。本项目以3D LiDAR点云为研究对象，充分挖掘道路曲面自身特征并引入鲁棒回归模型与样条平滑方法，系统建立一整套集“点云滤波”—“边界拟合”—“曲面平滑”等核心功能为一体的新方法，从根本上提高3D道路曲面重建的速度与质量。项目研究成果有助于促进LiDAR点云数据处理与三维地物重建研究的深入发展。

机载LiDAR技术可实现快速获取高精度地物三维坐标信息，在地表数据采集方面具有传统航空摄影测量所无法比拟的巨大优势。针对基于LiDAR点云的道路曲面重建中的点云滤噪声波不彻底，算法效率过低，重建曲面不够光滑等问题，该课题着重开展了以下研究：1）基于K-CNN深度学习模型的点云滤波方法，实现感兴趣点云与背景点云的二值化分类；2）基于降维映射与样条回归的三维道路曲面重建方法，系统建立了一整套的包含“点云滤波”—“边界拟合”—“曲面平滑”等核心功能为一体的技术流程，提高了3D道路曲面重建的速度与质量，可为加快数字化城市建设进程提供技术支持；3）基于道路曲率半径与建议行驶车速相吻合的匝道面生成方法，减弱或抵消弯道行车时的道路面的侧向摩擦力，保证车辆行驶安全；4）提出了一种基于张量分解与坡度自适应穿透率的机载LiDAR点云数据的断裂线自动提取算法，可以辅助运用于地震孕育机理解释。