摘要
将电网生产场景的三维数字化重建呈现,是提升物理电网可观可测能力的重要基础。传统采用激光雷达的三维重建存在采集工作耗时耗力问题,不适合大规模大面积的变电站三维重建。本文采用单/多目摄像头采集方式通过视觉三维重建技术构建变电站三维数字化模型,同时基于视觉三维重建结果开展变电站三维语义分割算法研究和增强现实AR智能巡检场景验证。通过定性和定量的实验分析,结果表明针对变电站基于视觉三维重建及语义级分割取得理想的效果,具有模型精度高、语义级分割准确率高,并成功验证了变电站AR智能巡检作业,为后续变电站数字化三维建设验证了技术可行性,对数字电网提升感知和决策具有重要意义。
The three-dimensional digital reconstruction of the power grid production scene is an important basis for improving the observability of the physical power grid. The traditional 3D reconstruction using lidar has the problem of time-consuming and labor-consuming in acquisition, which is not suitable for large-scale and large-area substation 3D reconstruction. In this paper, the single/multi camera acquisition method is used to construct the three-dimensional digital model of substation through visual three-dimensional reconstruction technology. At the same time, based on the visual three-dimensional reconstruction results, the research on the three-dimensional semantic segmentation algorithm of substation and the verification of augmented reality ar intelligent patrol scene are carried out. Through qualitative and quantitative experimental analysis, the results show that ideal results are achieved for substation based on visual 3D reconstruction and semantic level segmentation, with high model accuracy and semantic level segmentation accuracy, and the substation ar intelligent patrol inspection operation is successfully verified. It verifies the technical feasibility for the follow-up digital three-dimensional construction of substation, and is of great significance to improve the perception and decision-making of digital power grid.
作者
陈英达
曾群生
陈彬
黄文琦
梁凌宇
敖榜
CHEN Yingda;ZENG Qunsheng;CHEN Bin;HUANG Wenqi;LIANG Linyu;AO Bang(Digital Grid Research Institute Co.,Ltd.,Guangzhou 510700,Guangdong,China;China Southern Power Grid Co.,Ltd.,Guangzhou 510700,Guangdong,China)
出处
《电力大数据》
2022年第10期45-52,共8页
Power Systems and Big Data
关键词
视觉三维重建
三维语义分割
增强现实AR
智能巡检
visual 3D reconstruction
3D semantic segmentation
augmented reality
intelligent inspection
