LIDAR点云数据全自动滤波算法研究

An Automatic Point Clouds Filtering Algorithm Based on Grid Partition and Simplified Moving Least Squares

提出了一种基于移动最小二乘法的点云数据全自动滤波算法,该方法首先对LIDAR点云数据进行合理分块,并建立分块网格的动态四叉树空间索引,便于数据操作和管理.对分块网格中的点云数据利用精简移动最小二乘法拟合出参考地形,将拟合得到的参考地形用于LIDAR点云高程阈值的迭代计算,将每次迭代前后高差小于阈值的点划为地面点,其余点划分为非地面点,迭代运算直至阈值满足要求为止.实验表明,精简移动二乘法效率高,计算量小,并且精度高,适合点云数据DEM(digital elevation model)拟合,利用该算法对LIDAR点云数据进行滤波的速度快、精度高,能够有效地识别地面点和非地面点,并保留地形的细节信息.

An automatic point clouds filtering algorithm is presented on the basis of Grid Partition using Dynamic Quad Trees and reference surface fitted by Moving Least Squares. The filtering processing contains three major steps： Firstly,it gives the LIDAR point clouds reasonable grid partitions and establishes the corresponding dynamic quad trees spatial indices. Secondly,the points in the partitioned grids are utilized to fit a DEM reference plane using moving least squares technology. Finally,the elevation threshold is setup to separate ground points from those non-ground ones who are positioned above the reference plane and have a distance exceeding the threshold value to the plane. The aforementioned steps have to be repeated on the obtained ground points with gradually decreasing thresholds and grid size until desired precision is achieved. The experiments show that simplified moving least squares is high efficiency,small amount of calculation and high precision DEM data for point cloud fitting,and the filtering algorithm has high precision and can effectively identify ground points and non-ground ones without losing the detailed information of topography.