基于改进YOLOv8的高速公路区段车辆异常行驶事件研究

Research on Abnormal Driving Events of Vehicles in Highway Sections Based on Improved YOLOv8

摘要: 随着高速公路通车里程和交通流量的增加，车辆异常行驶事件引发的安全问题日益突出。传统监控手段存在效率低、漏检误检率高等不足，而基于深度学习的计算机视觉技术为解决这一问题提供了新途径。文章提出一种基于改进YOLOv8的高速公路车辆异常行为检测方法，通过引入Swin Transformer结构替换主干网络，增强模型对长距离特征和时空关联的建模能力。研究构建了包含400张图像的高速公路车辆行驶事件数据集，涵盖不同时段、天气和车辆类型。实验结果表明，改进后的模型在精确率(92.5%)、召回率(98%)和平均精确率(mAP0.5为94.4%)等指标上显著优于原始YOLOv8模型，能够有效提升高速公路车辆异常行为的检测精度和可靠性，为智能交通系统的实时监控与事故预防提供了技术支撑。

Abstract: With the increase in the mileage of expressways and traffic flow, safety issues caused by abnormal vehicle driving events have become increasingly prominent. Traditional monitoring methods suffer from low efficiency, high missed detection and false detection rates, while computer vision technology based on deep learning provides a new approach to solve this problem. This paper proposes an improved YOLOv8-based method for detecting abnormal vehicle behaviors on expressways by introducing the Swin Transformer structure to replace the backbone network, enhancing the model's ability to model long-range features and spatio-temporal correlations. A dataset of 400 images of vehicle driving events on expressways was constructed, covering different time periods, weather conditions, and vehicle types. Experimental results show that the improved model significantly outperforms the original YOLOv8 model in terms of precision (92.5%), recall (98%), and mean average precision (mAP 0.5: 94.4%), effectively improving the detection accuracy and reliability of abnormal vehicle behaviors on expressways. This study provides technical support for real-time monitoring and accident prevention in intelligent transportation systems.