摘要
道路两侧的路灯、交通标志杆和交通信号杆等杆状道路设施的位置与状态等信息是道路调查,以及城市部件管理中需要获取的重要数据。本文首先利用区域生长和渐进形态学滤波移除地面点,并对非地面点聚类;然后对聚类对象进行体素化网格采样,并提取采样点的快速点特征直方图,使用费舍尔编码对快速点特征直方图进行编码得到费舍尔向量;结合费舍尔向量以及地物的几何和辐射等全局特征,采用支持向量机(SVM)分类器进行训练和分类,实现杆状道路设施的识别和分类。实验结果表明,本文方法对路灯、交通标志杆和摄像头杆的提取准确率分别为89.3%、66.7%和100%,漏分率分别为10.7%、33.3%和0%,错分率分别为13.5%、4.8%和6.3%。
Information such as the location and status of rod-shaped road facilities such as street lights, traffic signs and traffic signal poles on both sides of the road is important data to be acquired in road surveys and urban items management. In this paper, we first use regional growth and progressive morphology filtering to remove ground points and cluster non-ground points. Then we perform voxelization on clustering objects and extract the fast point feature histogram of sampling points. Then we encode the fast point feature histogram to obtain the Fisher vector. Combined with the Fisher’s vector and the global features such as geometry and radiation, the SVM classifier is used for training and classification to identify the rod-shaped road facilities. The experimental results show that the extraction accuracy rates of streetlights, traffic signs and camera poles are 89.3%, 66.7% and 100%, respectively, the leak rate is 10.7%, 33.3% and 0%, respectively, and the error rate is 13.5%, 4.8% and 6.3% respectively.
作者
刘元元
Liu Yuanyuan(China Railway First Survey And Design Insitute Group Co.,Ld.,Xi'an 710043,China)
出处
《工程勘察》
2022年第4期62-66,共5页
Geotechnical Investigation & Surveying
关键词
车载激光点云
目标识别
杆状道路设施
费舍尔编码
分类
mobile laser point cloud
object recognition
rod-shaped road facilities
Fisher encoding
classification
