基于YOLOv8和ByteTrack的车辆检测和跟踪算法研究  

Research on Vehicle Detection and Tracking Algorithm Based on YOLOv8 and ByteTrack

作　　者：王佳丽 狄巨星[1] 杨阳[1] 刘贵锁 WANG Jiali;DI Juxing;YANG Yang;LIU Guisuo(Hebei Institute of Architecture and Civil Engineering,Zhangjiakou Hebei 075000)

机构地区：[1]河北建筑工程学院,河北张家口075000

出　　处：《长江信息通信》2024年第10期50-53,共4页Changjiang Information & Communications

基　　金：校级创新基金项目(XY2024042)。

摘　　要：运动车辆的检测和跟踪是智能交通系统的关键技术之一,传统的车辆检测和跟踪方法存在着实时性差,易受背景环境干扰、车辆形态相似导致车辆误检等情况。为了解决这个问题,提出基于YOLOv8和ByteTrack的车辆检测和跟踪算法。在车辆检测阶段,针对模型结构复杂、计算量大等问题,将YOLOv8的骨干网络替换为轻量级的网络MobileNetV3,以减少模型的参数量和计算量,保证车辆检测和跟踪的实时性;针对车辆在摄像头拍摄过程中存在误检的问题,将YOLOv8检测头替换为DyHead动态目标检测头,可以更精确地识别目标车辆。最后采用改进YOLOv8检测算法和ByteTrack跟踪算法结合来完成多目标车辆跟踪,经实验证明,该方法在保证精度几乎不变的情况下参数量降低了45.0%,计算量降低了46.3%,证明了算法的有效性,改进后的模型有较好的实时性与跟踪准确率,满足实际的使用需求。The detection and tracking of moving vehicles is one of the key technologies of intelli-gent transportation system,and the traditional vehicle detection and tracking mcthods have poor real-time performance,easy to be disturbed by the background environment,and the similarity of vehicle morphology leads to vehicle misdetection.In order to solve this problem,a vehicle de-tection and tracking algorithm based on YOLOv8 and ByteTrack is proposed.In the vehicle de-tection stage,for the problems of complex model structure and large computation,the backbone network of YOLOv8 is replaced by a lightweight network MobileNetV3 to reduce the number of paramcters and computation of the model and ensure the real-time vchicle detection and tracking;for the problems of vehicle misdetection during camera shooting,the YOLOv8 detec-tion head is replaced by the DyHead dynamic target detection head,which can recognize the tar-get vchicle more accuratcly.Finally,the improved YOLOv8 detection algorithm and ByteTrack tracking algorithm are combined to complete the multi-target vehicle tracking,and it is proved by the experiments that the number of parameters is reduced by 45.0%and the amount of com-putation is reduced by 46.3%while the accuracy is almost unchanged,which proves the effec-tiveness of the algorithm,and thc improved model has a better real-time performance and track-ing accuracy,which can satisfy the practical use requirements.

关 键 词：车辆检测 车辆跟踪 YOLOv8 ByteTrack 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]