基于区域生长的复杂场景点云分割  

Segmentation of complex scene point cloud based on region growing

作　　者：封功源 施卫[1] 李展峰 常嘉伟 刘斌[1] FENG Gongyuan;SHI Wei;LI Zhanfeng;CHANG Jiawei;LIU Bin(School of Automotive and Traffic Engineering,Jiangsu University of Technology,Changzhou 213000,China)

机构地区：[1]江苏理工学院汽车与交通工程学院,江苏常州213001

出　　处：《江苏理工学院学报》2023年第2期56-63,共8页Journal of Jiangsu University of Technology

基　　金：2019年江苏省社会科学基金项目“公共利益导向的产教联合体培养职教师资的路径与策略研究”(19JYB012)。

摘　　要：复杂场景下点云的有效分割是工程实际中所面临的难题,借助PCL点云库和KITTI数据集能实现对复杂场景下点云的有效分割。针对激光雷达扫描出来的点云数据具有无序性、散乱和不均匀的特点,在实现复杂场景下点云分割前,首先将点云进行地面点云与非地面点云的分割,再用下采样的方式简化复杂场景下的点云图像。通过估算点云曲率的大小,选取点云图像中曲率最小的点为种子点,从最平坦的区域进行分割。实验结果表明,算法对复杂场景下的点云分割精度高、准确性好。The effective segmentation of point clouds in complex scenes is a difficult problem in engineering practice.With the help of PCL point cloud library and KITTI data set,the effective segmentation of point clouds in complex scenes can be realized.The point cloud data scanned by laser radar has the characteristics of disorder,scattered and uneven.Before realizing the point cloud segmentation in complex scenes,the point cloud is first segmented into ground point cloud and non-ground point cloud,and then the point cloud image in complex scenes is simplified by down sampling.By estimating the curvature of the point cloud,the point with the smallest curvature in the point cloud image is selected as the seed point to segment from the flattest area.The experimental results show that the presented method has high precision and good accuracy for point cloud segmentation in complex scenes.

关 键 词：复杂场景 KITTI数据集 点云分割 区域生长 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]