期刊文献+

四旋翼无人机俯拍视角下的行人检测与轨迹追踪 被引量:2

Pedestrian detection and route tracking from aerial view of quad-rotor UAVs
下载PDF
导出
摘要 针对无人机拥有更宽广拍摄视角和更灵活拍摄姿态的优势,为了实现并定量地评估现有的目标检测算法在无人机俯拍视角下行人检测和轨迹追踪的应用,构建了一种四旋翼无人机俯拍视角下的行人检测与行人轨迹追踪算法。该算法采用YOLOv5作为目标检测模型,使用四旋翼无人机实时采集的视频数据作为分析测试数据。检测中首先训练YOLOv5,通过对检测结果的统计,对无人机拍摄时的水平距离、垂直高度、行人运动姿态等参量进行定量分析验证,并在该算法绘制锚框的基础上勾勒出行人运动的轨迹曲线,从而实现了视频画面中行人运动的路径追踪。对实际拍摄的视频数据进行行人检测与轨迹追踪的结果表明,该算法在实际应用时对无人机俯拍具有15~20 m的截止高度arctan3至arctan4的截止角度和约20 m的截止距离要求,但受行人运动姿态的影响较小。该算法与其他常规目标检测算法相比性能较优,能有望用于对拍摄视角要求更宽广和拍摄姿态要求更灵活的场合。因此,本文基于YOLOv5实现了无人机俯拍视角下的行人检测和运动路径追踪,并定量分析计算出该算法在行人检测时的截止高度和截止距离要求,这对于实际中应用该算法开展无人机侦测或救援工作有指导意义。 Aiming at the advantages of UAVs with wider shooting view and more flexible shooting angles and in order to realize and quantitatively evaluate the application of existing object detection algorithms in pedestrian detection and route tracking from aerial view,this paper constructed a pedestrian detection and routes tracking algorithm from aerial shooting view of quad-rotor UAVs.The algorithm adopts YOLOv5 as object detection model,and uses video data collected in practical situation as test data.Model YOLOv5 is trained at first,then through the statistics of the detectin results,the parameters of UAVs’video-shooting such as horizontal distance,vertical height and pedestrians’postures during UAVs’shooting are quantitatively analyzed and verified.Based on the anchor frame,the route curves of pedestrians’movement is outlined to realize path tracking.The results of pedestrian detection and trajectory tracking demonstrate that the algorithm has certain cut-off height of 15 to 20 meters,cut-off angle of arctan3 to arctan4 and cut-off distance of about 20 meters in practical application,but it is less affected by pedestrian postures.Compared with other conventional object detection algorithms,this algorithm has better performance and is expected to be used in situations where the shooting angle is required to be broader and the shooting position is required to be more flexible.In addition,the cut-off height and cut-off distance requirements of the algorithm,calculated by this paper in pedestrian detection quantitatively,has guiding significance for the practical application of the algorithm to carry out UAV detection or rescue.
作者 何飞麒 He Feiqi(School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China)
出处 《电子测量技术》 北大核心 2022年第10期50-56,共7页 Electronic Measurement Technology
关键词 无人机 俯拍视角 行人检测 深度学习网络 YOLOv5模型 unmanned aerial vehicle(UAV) aerial view pedestrian detection deep learning network YOLOv5 model
  • 相关文献

参考文献13

二级参考文献31

共引文献145

同被引文献6

引证文献2

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