一种多尺度拟合曲面的LiDAR数据建筑物脚点提取方法

Research on Building Extraction Method of LiDAR Data Based on Multi-Scale Fitted Surfaces

针对现有的建筑物点云提取方法,提出一种基于多尺度拟合曲面算法的LiDAR数据建筑物脚点提取方法。采用顾及地形的TIN滤波算法对地面点和非地面点进行分类;对非地面点集进行多尺度曲面拟合,提取建筑物候选点云;通过分析最大重叠边界矩阵MOBR和阈值条件(如距地高程和最小面积等)剔除具有类似几何和表面特性的人造物(如桥梁、车辆和防护堤等),实现建筑物点云的提取。使用国际摄影测量与遥感协会(ISPRS)提供的Vaihingen和Toronto两块测试数据进行评估,基于面积和目标的平均质量分别达到91.6%、93.1%、84.8%和86.5%。实验结果表明,该方法能够稳健地提取建筑物点云,具有较高的正确性和完整性。

In view of existing building extraction methods, a method for extracting building points of LiDAR data based on multi-scale fitting surface is proposed. The ground point and non-ground points are classified using the TIN filtering algorithm which takes into account the terrain. The non-ground pointset is multi-scale surface fitting to extract the building candidate point cloud. Eliminating the artifacts (bridges, vehicles, dikes, etc.) which has the same geometric and surface attributes as a building by analyzing RMBR and threshold conditions (elevation, minimum area, etc.) to achieve extraction of buildings. This method is evaluated using test data, Vaihingen and Toronto, provided by the International Society for Photogrammetry and Remote Sensing (ISPRS). The Results show the average quality of area-based and object-based are 91.6%, 93.1%, 84.8% and 86.5%. It demonstrates that the method can robustly extract building points with high correctness and integrity.