多尺度点云特征随机森林滤波算法

Multiscale point cloud feature random forest filtering algorithm

针对从大量点云数据中高效、便捷地提取地面点云的问题,提出一种耦合多尺度点云特征和随机森林模型的滤波算法(MFRF)。首先选取一部分试验区数据作为训练样本,采用人机交互方式将训练样本中地面点和植被点进行分类并标记。然后将点云RGB信息转换为HSV颜色信息,采用主成分分析法计算出多尺度下点云特征值。最后将带有标签、颜色信息和特征值的训练样本放入随机森林分类器中进行训练,将构建的随机森林分类器应用到待分类点云上,进行点云滤波。该算法能有效地分离出地面点与植被点,较为完整地保留了地面点云。将MFRF与CSF滤波算法、坡度滤波算法、形态学滤波算法进行对比分析,结果表明该算法优于对比算法,试验区A、B滤波精度分别提高5.12%和6.89%,验证了该算法的有效性、适用性。

To extract ground point cloud efficiently and conveniently from a large number of point cloud data, a filtering algorithm combining multi-scale point cloud features with random forest model(MFRF) was proposed. Firstly,some test area data were selected as training samples, and ground points and vegetation points in the training samples were classified and marked by human-computer interaction. Then, the RGB information of point cloud is converted into HSV color information, and the characteristic values of point cloud at multi-scale are calculated by principal component analysis. Finally, the training samples with labels, color information and eigenvalues are put into the random forest classifier for training, and the constructed random forest classifier is applied to the point cloud to be classified for point cloud filtering. The algorithm can effectively separate ground points from vegetation points and preserve the ground point cloud completely. MFRF was compared with CSF filtering algorithm, slope filtering algorithm and morphology filtering algorithm. The results show that the algorithm is superior to the comparison algorithm, and the filtering accuracy of test area A and B is improved by 5. 12% and 6. 89% respectively, which verifies the effectiveness and applicability of the algorithm.