多目标跟踪技术在隧道复杂交通环境中的应用

Application of Multi Object Tracking Technology in Complex Traffic Environment of Tunnels

摘要: 针对隧道复杂交通环境中多目标车辆跟踪存在的遮挡频繁、光照骤变等问题，本文对，基于YOLOv5 + ByteTrack的多目标跟踪技术。通过分析隧道环境特性，设计了包含目标检测与跟踪关联的算法框架，采用YOLOv5作为目标检测器，ByteTrack作为跟踪算法。实验结果表明，在遮挡与车流密度极端情况下，YOLOv5 + ByteTrack的ID保持率(95.7%)显著高于YOLOv5 + DeepSORT (71.4%)；在车辆速度变化与运动模糊影响下，ByteTrack算法对图像模糊的依赖性更小，追踪连续性表现更优。该技术为隧道交通监控提供了有效解决方案，对提升隧道交通安全监测与异常行为识别能力、推动隧道交通向自动化、智能化管控转型具有重要现实意义。

Abstract: To address issues such as frequent occlusions and sudden changes in illumination in multi-object vehicle tracking within complex tunnel traffic environments, this paper focuses on multi-object tracking technology based on YOLOv5 ByteTrack. By analysing the characteristics of tunnel environments, an algorithm framework incorporating object detection and tracking association was designed, using YOLOv5 as the object detector and ByteTrack as the tracking algorithm. Experimental results show that under extreme conditions of occlusion and traffic density, the ID retention rate of YOLOv5 ByteTrack (95.7%) is significantly higher than that of YOLOv5 DeepSORT (71.4%); under the influence of vehicle speed changes and motion blur, the ByteTrack algorithm is less dependent on image clarity and exhibits better tracking continuity. This technology provides an effective solution for tunnel traffic monitoring and is of significant practical importance for enhancing tunnel traffic safety monitoring, recognising abnormal behaviours, and promoting the transition of tunnel traffic towards automated and intelligent management.