摘要
针对配电房点云场景的三维重建需求问题,该文提出了一种配电房结构及设备点云的自动化语义分割方法。对于配电房主体结构点云,先采用区域增长的方法获取所有结构面的点云块,再基于随机采样一致性平面拟合的方法,对属于同一结构面的点云进行精准合并;对配电房内部设备点云,先采用欧式聚类的方法对每个设备点云进行聚类分割,再采用点云投影与栅格化的方法生成每个设备的主面图像,最后基于深度残差网络搭建深度神经网络对每个设备主面图像进行分类,得到每个设备点云的语义信息。实验证明,该文提出的方法对配电房点云分割的正确率优于0.9,设备识别的正确率优于96%,能够满足后续配电房及设备三维重建的要求。
In order to meet the demand of three-dimensional scene reconstruction of point cloud in power distribution room,an automatic semantic segmentation method for point cloud of power distribution room structure and equipment was proposed based on the geometric characteristics.For the point cloud of the main structure of each power distribution room,the method of regional growth was first adopted to obtain the point cloud blocks of all structural planes.And then,the point clouds belonging to the same structural plane were accurately combined based on the random sample consensus plane fitting method.For the equipment point cloud,the European clustering method was first used to segment each equipment point cloud.And then,the point cloud projection were used to generate the main surface image of each equipment.Finally,a deep neural network was built based on the deep residual networks to classify the main surface image of each equipment for obtaining the semantic information of each equipment.Experiment results showed that the accuracy of the proposed method for point cloud segmentation was better than 0.9,and the accuracy of equipment classification was better than 96%,which could meet the requirement of three-dimensional reconstruction of the distribution room and equipment.
作者
金文佩
梁倩仪
于安斌
杨叶铠
梅文胜
JIN Wenpei;LIANG Qianyi;YU Anbin;YANG Yekai;MEI Wensheng(Guangzhou Liwan Power Supply Bureau,Guangdong Power Grid Co.,Ltd.,Guangzhou 510000,China;School of Geodesy and Geomatics,Wuhan University,Wuhan430079,China)
出处
《测绘科学》
CSCD
北大核心
2022年第12期184-191,199,共9页
Science of Surveying and Mapping
基金
国家自然科学基金项目(41674005)
南方电网有限责任公司科技项目(GZHKJXM20180145)
关键词
配电房
三维激光点云
语义分割
区域增长
欧式聚类
distribution room
three dimensional laser point cloud
point cloud segmentation
regional growth
euclidean clustering
作者简介
金文佩(1990—),女,广东广州人,工程师,硕士,主要研究方向为配网规划及应用。E-mail:991740191@qq.com;通信作者:梅文胜,教授,E-mail:wshmei@sgg.whu.edu.cn
