期刊文献+

基于目标检测与跟踪的轻量化煤矿区域限员算法

A lightweight algorithm for coal mine regional personnel restriction based on object detection and tracking
在线阅读 下载PDF
导出
摘要 目前基于视频监控的井下区域限员算法主要利用多目标跟踪实现,存在运算复杂难以实现低功耗嵌入式移植,易受环境光源干扰等问题。针对上述问题,提出了一种基于目标检测与跟踪的轻量化煤矿区域限员算法。利用多台双目结构光摄像仪采集重点区域进出口的彩色图和深度图,基于改进的轻量化目标检测模型GF−YOLOv8,完成工作人员的图像定位;采用交并比优先策略对目标与轨迹进行匹配,减少了传统目标跟踪算法的匹配次数,提升了跟踪的运算速度;以跟踪轨迹为基础,通过多摄像仪的跨线计数的方式,统计重点区域总人数,当达到限员人数时,输出报警信息。试验结果表明:双目结构光摄像仪可以有效克服光照影响;改进的目标检测模型GF−YOLOv8的AP@0.5和AP@0.5:0.95较YOLOv8分别提高了0.24%,3.42%,参数量和每秒浮点运算次数分别降低了17.29%,2.58%;交并比优先策略与DeepSort算法精度相近,但具有更低的匹配复杂度;在限员报警次数方面,真实报警次数为17次,交并比优先策略正确报警16次,漏报1次。 At present,underground regional personnel restriction algorithms based on video monitoring mainly rely on multi-object tracking,which has problems such as complex computation that makes low-power embedded deployment difficult and susceptibility to environmental light interference.To address the above issues,a lightweight algorithm for coal mine regional personnel restriction based on object detection and tracking was proposed.Multiple binocular structured-light cameras were used to collect color and depth images at key entrances and exits.Based on the improved lightweight object detection model GF-YOLOv8,worker image localization was achieved.An Intersection-Over-Union priority strategy was adopted to match objects and trajectories,which reduced the number of matches compared with traditional tracking algorithms and improved the computational speed of tracking.Based on tracking trajectories,the total number of people in key areas was counted through the cross-line counting method with multiple cameras.When the number of people in the area reached the personnel limit,alarm information was generated.The experimental results showed that binocular structured-light cameras effectively overcame the influence of illumination.The AP@0.5 and AP@0.5:0.95 of the improved detection model GF-YOLOv8 increased by 0.24%and 3.42%respectively compared with YOLOv8,while the number of parameters and floating point operations per second decreased by 17.29%and 2.58%respectively.The Intersection-Over-Union priority strategy achieved accuracy close to the DeepSort algorithm but with lower matching complexity.Regarding the count of personnel restriction alarms,there were 17 true alarms;the Intersection-Over-Union priority strategy correctly reported 16 and missed 1 alarm.
作者 樊荣 邓飞 游磊 杨筱彧 李涛 FAN Rong;DENG Fei;YOU Lei;YANG Xiaoyu;LI Tao(CCTEG Chongqing Research Institute,Chongqing 400037,China;State Key Laboratory of Coal Mine Disaster Prevention and Control,Chongqing 400037,China)
出处 《工矿自动化》 北大核心 2025年第8期51-58,共8页 Journal of Mine Automation
基金 重庆市自然科学基金博士直通车项目(CSTB2024NSCQ-BSX006) 中煤科工集团重庆研究院有限公司一般项目(2025YBXM06)。
关键词 煤矿区域限员 区域人数统计 目标检测与跟踪 双目结构光摄像仪 GF−YOLOv8 交并比优先策略 coal mine personnel restriction regional population statistics object detection and tracking binocular structured-light camera GF-YOLOv8 Intersection-Over-Union priority strategy
  • 相关文献

参考文献7

二级参考文献54

共引文献22

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部