基于轻量型YOLOv5和DeepSort的跟车辅助预警算法  

Vehicle Tracking Assisted Warning Algorithm Based on Lightweight YOLOv5 and DeepSort

作　　者：张海川 ZHANG Haichuan(School of Electronic and Electrical Engineering,Chongqing University of Science and Technology,Chongqing 401331,China)

机构地区：[1]重庆科技大学电子与电气工程学院,重庆401331

出　　处：《现代信息科技》2024年第18期59-65,共7页Modern Information Technology

摘　　要：针对当前主流的车辆驾驶辅助算法对硬件算力依赖较强的问题,提出了一种基于轻量型YOLOv5和DeepSort的车辆跟车辅助预警算法。该算法模型包括目标检测、目标追踪和跟车预警三个模块,能够实时检测道路前方的车辆、追踪并测量车距和车速,根据前车的车距和车速情况判断风险等级并做出预警。该算法在耗时最长的目标检测模块部分选择检测速度和精度都很优秀的YOLOv5s并对其做轻量化处理,以Mobilenetv3代替原YOLOv5s的backbone骨架并在特征融合层引入GSConv卷积,此外针对原CIoU损失函数在该算法中收敛效果不理想的问题,引入了WIoU损失函数。实验结果表明,与直接使用YOLOv5s作为检测模块相比,算法的参数量下降了37%,权重文件大小下降36.1%,检测速度提升了23.8%,mAP值仅下降0.3%。In view of the problem that the current mainstream vehicle driving assisted algorithm has strong dependence on hardware computing power,a vehicle tracking assisted warning algorithm based on lightweight YOLOv5 and DeepSort is proposed.The algorithm model includes three modules of target detection,target tracking and vehicle tracking warning.It could detect the vehicles ahead of the road in real time,track and measure the distance and speed,and judge the risk level according to the front car distance and speed and make early warning.The algorithm selects YOLOv5s with excellent detection speed and accuracy in the longest time-consuming target detection module and makes lightweight processing,replaces Mobilenetv3 with the backbone skeleton of the original YOLOv5s and introduces GSConv convolution in the feature fusion layer.In addition,aiming at the problem that the convergence effect of the original CIoU loss function is not ideal in the algorithm,the WIoU loss function is introduced.The experimental results show that the number of parameters of the algorithm has decreased by 37%,the weight file size has decreased by 36.1%,the detection speed has increased by 23.8%and the mAP value has decreased by only 0.3%compared with directly using YOLOv5s as the detection module.

关 键 词：车辆检测 目标追踪 神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.4[自动化与计算机技术—控制科学与工程]