基于Otsu的建筑物点云分割改进算法

Improved Algorithm of Building Point Cloud Segmentation Based on Otsu

提出了一种基于Otsu的建筑物点云分割改进算法,针对传统Otsu分割算法在精确性方面的不足提出2点修改意见。首先根据投影点密度理论将三维点云转换为二维灰度图像,再利用Otsu算法及其改进算法对图像进行分割。采用Riegl公司VZ-400激光扫描仪采集的武汉大学信息学部第一教学楼的点云进行分割处理,并与传统的Otsu分割算法进行对比。结果表明,该算法分割正确率可达96.54%,远高于改进前的72.82%。

In view of the defect of traditional Otsu algorithm in building point cloud segmentation, this paper proposed an improved algorithm. Firstly, 3D point cloud was transformed into 2D image according to the theory of projection point density. Then, two-dimensional image was segmented by traditional and improved Otsu algorithm. The algorithm proposed in this paper was tested by the building point cloud of Wuhan University Information Faculty Number 1 building by vz-400 LiDAR scanner. The accuracy was enhanced significantly from 72.82% to 96.54% compare to unimproved method.