基于YOLOv4与改进DeepSORT算法的车流量检测  

Traffic Flow Detection Based on YOLOv4 and Improved DeepSORT Algorithm

作　　者：王继超 张丽娟 张春茜 回振桥 申耀辉 WANG Ji-chao;ZHANG Li-juan;ZHANG Chun-qian;HUI Zhen-qiao;Shen Yao-hui(Department of Electrical Automation,Hebei University of Water Resources and Electric Engineering,061001,Cangzhou,Hebei,China;Water Resources Automation and Informatization Application Technology and Development Center of Hebei Colleges,061001,Cangzhou,Hebei,China)

机构地区：[1]河北水利电力学院电气自动化系,河北省沧州市061001 [2]河北省高校水利自动化与信息化应用技术研发中心,河北省沧州市061001

出　　处：《河北水利电力学院学报》2023年第1期6-11,共6页Journal of Hebei University Of Water Resources And Electric Engineering

基　　金：河北省教育厅科学技术研究项目(QN2021228);河北省大学生创新创业训练计划(S202010085030);沧州市重点研发计划指导项目(204102010);沧州市重点研发计划指导项目(204102014);河北省教育厅科学技术研究项目(ZC2022018)。

摘　　要：为了提高不同时段车流量的检测率,本文在优良的目标检测模型YOLOv4基础上,将传统的DeepSORT算法进行改进,将原有的IoU变为CIoU,保留追踪信息的同时,提供了更丰富匹配策略,使得目标追踪更加稳定,在一定程度上解决了光线较暗容易丢失目标的问题。最后在车流量检测阶段将本文改进算法与SORT算法、DeepSORT算法进行对比试验。试验结果表明,在白天情况下,本文算法相较于SORT算法提高10.7%,较DeepSORT算法提高1.3%;在夜晚情况下,本文算法相较于SORT算法提高18.1%,较DeepSORT算法提高7.9%。本文改进算法在夜晚车流量检测精准,从而为环境昏暗的条件下物体目标检测与追踪提供了理论参考与方法依据。In order to improve the detection rate of traffic flow in different time periods,based on the excellent target detection model YOLOv4,this paper improved the traditional DeepSORT algorithm,changed the original IoU into CIoU,and provided more abundant matching strategies to make the target tracking more stable while retaining the tracking information.To a certain extent,it solves the problem that the target is easily lost when the light is dim.Finally,the improved algorithm is compared with SORT algorithm and DeepSORT algorithm in traffic flow detection stage.Experimental results show that the proposed algorithm improves by 10.7%compared with SORT algorithm and 1.3%compared with DeepSORT algorithm in daytime.At night,the proposed algorithm improves 18.1%compared with SORT algorithm and 7.9%compared with DeepSORT algorithm.In this paper,the improved algorithm is accurate in detecting traffic flow at night,which provides theoretical reference and method basis for object detection and tracking in dim environment.

关 键 词：车流量检测 YOLOv4 DeepSORT 目标追踪 

分 类 号：U491[交通运输工程—交通运输规划与管理] TP391.41[交通运输工程—道路与铁道工程]