基于局部关系特征和注意力机制的交通事故预测方法  

Traffic Accident Prediction Method Based on Local Relational Features and Attention Mechanisms

作　　者：张亚辉 李颖 刘天恩 ZHANG Yahui;LI Ying;LIU Tianen(School of Mechanical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;School of Mechanical Engineering,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]燕山大学机械工程学院,河北秦皇岛066004 [2]北京理工大学机械与车辆学院,北京100081

出　　处：《北京理工大学学报》2025年第1期11-18,共8页Transactions of Beijing Institute of Technology

基　　金：国家自然科学基金资助项目(52102449);中国博士后基金资助项目(2021M690394)。

摘　　要：基于相机的事故预测方法主要是建立交通对象的全局关系,而缺乏对局部关系的考虑.为此,提出一种基于局部关系特征和注意力机制的预测模型,实现车载相机实时预测交通事故风险.该模型首先引入局部关系多图网络,捕捉车辆间的局部交互关系,解决交通对象的局部交互信息应用不充分的问题.其次利用动态空间注意力机制确定交通事故风险车辆.最后将门控循环网络和动态时间注意力机制联合,有效利用动态场景当前帧和历史帧间的时序信息.事故数据集的实验结果表明,该模型准确率达到73.78%,提前1.55s预测交通事故风险,同时单帧预测时间为1.65 ms,具有卓越的实时性.为交通事故风险预测提供一种有效的解决方案.The accident prediction methods with cameras are mainly used to establish the global relationships among traffic objects,lacking the consideration of local relationships among them.Therefore,a predictive mod-el was proposed based on local relational features and attention mechanisms to carry out real-time traffic acci-dent risk prediction with vehicle-mounted cameras.Firstly,a local relational multi-graph network was incorpor-ated to capture the local interactions of vehicles,solving the insufficient application issue of local interaction in-formation about the traffic objects.Subsequently,a dynamic spatial attention mechanism was adopted to identify the risk vehicles at traffic accident.Finally,the Gated Recurrent Unit and dynamic temporal attention mechan-ism were integrated to effectively utilize the temporal information between the current and historical frames in dynamic scenes.Experimental results on the accident dataset show that the proposed model can predict accident in 1.55 seconds advance with a prediction accuracy of 73.78%and 1.65 ms single-frame prediction time,realiz-ing excellent real-time effectiveness,and providing an effective solution for the traffic accident prediction.

关 键 词：交通事故预测 图神经网络 注意力机制 

分 类 号：U461[机械工程—车辆工程]