摘要
为解决电力道路场景分割过程中,最终分割结果平均交并比(mIoU)较高问题,提出基于实景LiDAR点云数据的电力道路场景分割方法。采用虚拟网格保存离散点云数据,结合数学形态学滤波原理,获取预处理后的点云数据。依托于移动最小二乘法推算出法矢估计结果,根据显著性指标判别出特征点,最终提取出点云数据轮廓特征。以实景LiDAR点云数据为基础,在考虑轮廓特征的情况下生成保留边界的超体素结构,再应用自上而下的聚类分割算法(P-Linkage)即可得到场景分割结果。实验结果表明,新研究方法应用后,电力道路场景分割结果mIoU值大于70%,极大提升了道路场景分割质量。
In order to solve the problem of high average intersection union ratio(mIoU)of the final segmentation result in the process of power road scene segmentation,a power road scene segmentation method based on real-world LiDAR point cloud data was proposed.The virtual grid was used to store the discrete point cloud data,and the preprocessed point cloud data was obtained by combining the mathematical morphological filtering principle.Relying on the moving least squares method,the vector estimation results were calculated,the feature points were identified according to the significance index,and the contour features of the point cloud data were finally extracted.Based on the real-world LiDAR point cloud data,the super-voxel structure with boundary preservation was generated under the condition of considering the contour features,and then the top-down clustering and segmentation algorithm(P-Linkage)was applied to obtain the scene segmentation results.The experimental results showed that after the application of the new research method,the mIoU value of the power road scene segmentation result was greater than 70%,greatly improving the quality of road scene segmentation.
作者
周敬余
张宇潇
周子雅
张俊杰
李鑫卓
ZHOU Jingyu;ZHANG Yuxiao;ZHOU Ziya;ZHANG Junjie;LI Xinzhuo(Tongren Power Supply Bureau of Guizhou Power Grid Co.,Ltd.,Tongren 554300,Guizhou China;lectric Power Science Research Institute of Guizhou Power Grid Co.,Ltd.,Guiyang 550002,China)
出处
《粘接》
CAS
2024年第5期153-156,共4页
Adhesion
关键词
激光扫描测量
点云数据
滤波
聚类分割
电力道路场景
laser scanning measurement
point cloud data
filtering
cluster segmentation
electricity road scenarios
