基于UWB与视觉融合的储煤场目标检测与定位方法研究  

Multi-sensor fusion-based intelligent supervision system for coal storage yard operation safety

作　　者：朱艳通 刘鹏 王永波 杨鹏 李传祥 范丽[2] 王紫江 ZHU Yantong;LIU Peng;WANG Yongbo;YANG Peng;LI Chuanxiang;FAN Li;WANG Zijiang(Guoneng Ningxia Lingwu Power Generation Co.,Ltd.,Yinchuan 751400,China;Huzhou Research Institute of Zhejiang University,Huzhou 313299,China)

机构地区：[1]国能宁夏灵武发电有限公司,宁夏银川751400 [2]浙江大学湖州研究院,浙江湖州313299

出　　处：《煤炭科学技术》2024年第S2期331-342,共12页Coal Science and Technology

摘　　要：作业工人和车辆是煤场安全管理的关键对象,实时获得高精度定位信息对避免安全事故具有十分重要的意义。研究基于超宽带与视觉融合的人员车辆检测与定位方法,根据封闭煤场的场景特点,提出了基于飞行时间-时间到达差混合定位的方法,实现煤场内携带标签目标的高精度定位;为解决因遮挡导致超宽带定位中断的问题,采用视觉位置估计辅助超宽带定位,提出了基于Trans-former和Yolov7融合目标检测识别方法,根据目标在图像中的位置变化,估计目标的位置坐标,采用扩展卡尔曼滤波方法实现融合定位。该方法在国能宁夏灵武发电厂煤场进行测试,通过收集煤场内作业场景数据图像,制作了作业人员与车辆数据集,目标检测模型的平均准确率指标mAP@0.5为0.925,模型在现场部署后的图片检测速度为30fps。检测模型的试验结果表明该模型对储煤场中的主要目标具有较高的检测精度和较快的检测速度,实时性达到了煤场应用落地的实际需求。对比试验结果表明,提出的目标检测模型在不牺牲处理帧率的情况下,mAP@0.5相比于原始CNN网络Yolov7提升了1.4%。煤场内定位精度范围为0.5~0.9m,定位刷新频率为10fps,有效地解决因遮挡导致的UWB定位中断问题。Workers and vehicles are the key objects of coal yard safety management.It is of great significance to obtain high-precision positioning information in real time to avoid safety accidents.Studying the detection and positioning method of personnel and vehicle based on ultra-WB and visual integration,according to the scene characteristics of the closed coal yard,the method based on flight time-time arrival difference is proposed to realize the high-precision positioning of label targets in the coal yard;In order to solve the interruption of UWB location caused by occlusion,UWB detection and identification method based on Transformer and Yolov7 is proposed to estimate the position in the image,and the Kalman filter method is used to realize the fusion positioning.This method was tested in the coal yard of Ningxia Lingwu Power Plant of National Energy.By collecting the data images of operation scenes in the coal field,the data set of operators and vehicles was made.The average accuracy index of the target detection model mAP@0.5 is 0.925,and the image detection speed of the model deployed on the site is 30 frames per second.The experimental results of the detection model show that the model has high detection accuracy and fast detection speed for the main motion targets in the coal storage field,and meets the actual requirements of the coal field application in real time.The comparative experimental results show that the proposed target detection model improves the mAP@0.5 by 1.4% over the original Yolov7 without sacrificing the processing frame rate.The positioning accuracy range in the coal yard is 0.5~0.9 m,and the positioning refresh frequency is 10 frames per second,which effectively solves the problem of UWB positioning interruption caused by occlusion.

关 键 词：目标检测 超宽带定位 卷积神经网络 飞行时间 时间到达差 扩展卡尔曼滤波 

分 类 号：TD713[矿业工程—矿井通风与安全]