基于计算机视觉的室内5G pRRU站址标注工具  

Site Survey and Labeling System for Indoor 5G pRRU Based on Computer Vision

作　　者：杨凌霄 蒋品 曾丹[1] Yang Lingxiao

机构地区：[1]上海大学通信与信息工程学院,上海200444 [2]上海华为技术有限公司无线应用场景实验室,上海201206

出　　处：《工业控制计算机》2023年第12期1-3,共3页Industrial Control Computer

摘　　要：为了解决室内5G pRRU的站址测量不准确问题,提出了一种基于计算机视觉的pRRU站址标注工具。该工具首先使用双目相机采集pRRU图像,通过YOLOv4目标检测算法检测出pRRU,之后经过SGBM算法获取视差,最后通过坐标变换把pRRU的相机坐标系坐标映射到场景的世界坐标系中,从而实现对pRRU的精确识别和定位。实验结果表明,该工具对pRRU的检测精度达到99.8%,在场景中的定位最大误差为0.3 m,实时性方面在Jetson Nano上平均帧率达到27 fps。In order to solve the inaccurate site measurement problem of indoor 5G pRRU,a computer vision based pRRU site labeling tool is proposed in this paper.Firstly,the tool acquires pRRU images using a binocular camera,and detects pRRU by the YOLOv4 object detection algorithm.Then the disparity is obtained by SGBM algorithm.Finally,the camera coordinate system coordinates of pRRU are mapped to the world coordinate system of the scene through coordinate transformation,so as to achieve accurate identification and localization of pRRU.The experimental results show that the tool achieves 99.8%detection accuracy for pRRU and a maximum error of 0.3 m in localization in the scene,while the average video transmission frame rate reaches 27 fps on Jetson Nano.

关 键 词：pRRU 目标检测 深度测量 坐标映射 端云协同 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN929.5[自动化与计算机技术—计算机科学与技术]