基于监督学习的机载LiDAR点云滤波方法

Airborne LiDAR point cloud filtering method based on supervised learning

点云滤波是机载LiDAR点云后处理应用必不可少的环节。虽然国内外学者提出了许多滤波方法,但依然存在以下问题:一是需要复杂的参数设定;二是滤波方法鲁棒性较差,难以适应复杂的地形环境。为了实现高精度自动点云滤波,本文提出一种基于支持向量机(SVM)的滤波方法。该方法将点云滤波转换为二分类问题,通过计算点云数据的5维特征向量,使用计算出的模型将点云数据分为地面点和地物点两大类。本文采用国际摄影测量与遥感学会(ISPRS)网站发布的三组点云数据进行实验,实验结果表明该方法对具有不同地形特征的点云数据均能获得良好的滤波结果。在与其他三种代表性的滤波方法对比中,本文方法能够取得最小的滤波Ⅰ类误差(3.07%)和总误差(3.68%)。

Point cloud filtering is an essential step in the post-processing of airborne LiDAR point cloud. Although many filtering methods have been proposed by scholars at home and abroad, the following problems still exist: first, complex parameter setting is required;secondly, the filtering method has poor robustness and is difficult to adapt to complex terrain environment. In order to realize high-precision automatic point cloud filtering, a filtering method based on support vector machine(SVM) is proposed in this paper. In this method, point cloud filtering is transformed into a two-category problem, and the point cloud data are divided into ground point and ground object point by using the calculated model based on calculating the 5-dimensional feature vector of point cloud data. In this paper, three sets of point cloud data published on the website of ISPRS are used for experiments, and the experimental results show that this method can obtain good filtering results for point cloud data with different topographic features.Compared with other three representative filtering methods, the proposed method can achieve the minimum filtering class I error(3.07%) and total error(3.68%).