基于改进YOLOv6模型的交通异常事件检测算法研究  

Research on Traffic Anomaly Event Detection Algorithm Based on Improved YOLOv6 Model

作　　者：薛盘芬 靳凯斌 XUE Panfen;JIN Kaibin(Anhui Provincial Transportation Comprehensive Law Enforcement and Supervision Bureau,Hefei 230051;Chang′an University,Xi′an 710064)

机构地区：[1]安徽省交通运输综合执法监督局,合肥230051 [2]长安大学,西安710064

出　　处：《公路交通技术》2025年第1期176-181,189,共7页Technology of Highway and Transport

基　　金：中央高校基金项目(300102343514)。

摘　　要：交通异常事件检测是智能交通系统中的关键任务,但现有目标检测算法在该领域的应用尚存在技术瓶颈,针对检测精度不足、模型对复杂场景的适应性差以及缺乏高质量的公开数据集等问题,提出了一种改进的YOLOv6模型,旨在提高交通异常事件(如车辆碰撞、单车冲撞和车辆起火)检测的准确性和性能。先将原YOLOv6模型中的损失函数替换为CIoU损失函数,以增强模型的定位精度,后引入CBAM注意力机制,以提高模型对关键特征的关注度,再采用自动混合精度训练策略优化训练过程,最后为了验证改进效果,通过游戏引擎Grand Theft Auto V生成数据集,并对其进行标注,涵盖3类交通异常事件。试验结果表明:1)提出的改进YOLOv6模型在交通异常事件的检测任务中可获得87.2%的平均检测精度,在各项指标上表现优异;2)召回率AR较次优模型提高2.1%,IoU阈值为0.5时,平均精度mAP高出2.6%;IoU阈值为0.5至0.95时,mAP增长3.7%;3)车辆相撞、单车相撞和车辆起火烧毁的精度分别达到79.9%、37.6%和65.6%,均优于次优模型,验证了改进方法的有效性。Traffic anomaly event detection is a critical task in intelligent transportation systems,yet the application of existing object detection algorithms in this field still faces technical bottlenecks.Addressing issues such as insufficient detection accuracy,poor adaptability of models to complex scenes,and the lack of high-quality public datasets,an improved YOLOv6 model is proposed,aiming to enhance the accuracy and performance of traffic anomaly event detection(such as vehicle collisions,single-vehicle crashes,and vehicle fires).Initially,the loss function in the original YOLOv6 model is replaced with the CIoU loss function to improve the model′s localization accuracy.Subsequently,the CBAM attention mechanism is introduced to increase the model′s focus on key features.Then,an automatic mixed precision training strategy is adopted to optimize the training process.Finally,to validate the improvements,a dataset is generated using the game engine Grand Theft Auto V and annotated,covering three types of traffic anomaly events.Experimental results show that:1)The proposed improved YOLOv6 model achieves an average detection accuracy of 87.2%in the task of detecting traffic anomaly events,with excellent performance across various metrics;2)The recall rate AR is improved by 2.1%compared to the next best model,with the mean average precision(mAP)increasing by 2.6%at an IoU threshold of 0.5,and by 3.7%when the IoU threshold ranges from 0.5 to 0.95;3)The precision rates for vehicle collisions,single-vehicle crashes,and vehicle fires reach 79.9%,37.6%,and 65.6%respectively,all outperforming the next best model,thereby validating the effectiveness of the proposed improvements.

关 键 词：智能交通 交通异常事件检测 YOLOv6 CIoU损失函数 CBAM注意力机制 

分 类 号：U495[交通运输工程—交通运输规划与管理]