摘要
多目标跟踪(Multi-Object Tracking,MOT)在自动驾驶和智能监控等领域具有广泛应用。针对传统的行人检测与跟踪方法存在复杂度高、人群密集时易漏检以及遮挡影响严重等问题,提出一种基于改进YOLOv11和ByteTrack算法的行人检测与跟踪方法。在行人检测阶段,将标准卷积(Convolution,Conv)模块替换为ADown模块,以降低算法复杂度;在C2PSA模块中引入多尺度注意力(Efficient Multi-scale Attention,EMA)机制,进一步降低计算开销的同时,保留丰富的通道信息;采用动态目标检测头(DynamicHead,DyHead)提升行人检测精度,缓解摄像头拍摄场景下的漏检问题。在行人跟踪阶段,将ByteTrack算法与改进的YOLOv11算法结合,以实现鲁棒的行人跟踪。改进后的YOLOv11模型在mAP50提高2.2%,参数量减少25.2%,在复杂度与准确性方面能够满足实际应用需求。
Multi-Object Tracking(MOT)has extensive applications in fields such as autonomous driving and intelligent surveillance.Aiming at the problems of high complexity,frequent missed detection in dense crowds,and severe occlusion effects in traditional pedestrian detection and tracking methods,this paper proposes a pedestrian detection and tracking approach based on improved YOLOv11 and ByteTrack algorithms.In the pedestrian detection stage,the standard Convolution(Conv)module is replaced with the ADown module to reduce algorithm complexity.The Efficient Multi-scale Attention(EMA)mechanism is introduced into the C2PSA module to further reduce computational overhead while preserving rich channel information.A DynamicHead(DyHead)is adopted to enhance pedestrian detection accuracy and mitigate missed detection issues in camera capture scenarios.In the pedestrian tracking stage,the ByteTrack algorithm is integrated with the improved YOLOv11 to achieve robust pedestrian tracking.The modified YOLOv11 model demonstrates a 2.2%improvement in mAP50 and a 25.2%reduction in model parameters,effectively balancing complexity and accuracy to meet practical application requirements.
作者
刘兆金
谭钦红
朱嘉浩
李会兵
LIU Zhaojin;TAN Qinhong;ZHU Jiahao;LI Huibing(School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《激光杂志》
北大核心
2025年第10期56-62,共7页
Laser Journal
基金
国家自然科学基金(No.61771085)。
