摘要
基于宏观交通特征进一步理解交通冲突机制,提出一种基于监控视频的隧道交通冲突预测方法,以实现对隧道交通安全的风险预警.通过在车辆轨迹数据中检测出交通冲突事件及对应的宏观交通特征,使用二分类模型验证支持向量机、决策树、多层感知机和随机森林等模型预测交通冲突的可行性,建立交通状态与交通冲突的关联关系.研究结果表明:该预测方法使用的交通状态变量对交通冲突的发生概率均有显著性贡献.随机森林模型能够根据20 s时间单元采集的交通状态参数有效预测交通冲突的发生风险,预测准确率可达到97%.
To gain a deeper understanding of the traffic conflict mechanism based on macroscopic traffic characteristics,this paper proposes a tunnel traffic conflict prediction method based on surveillance video to provide risk warnings for tunnel traffic safety.By detecting traffic conflict events and corresponding macroscopic traffic features in the vehicle trajectory data,a binary classification model is used to verify the feasibility of traffic conflict prediction using support vector machines,decision trees,multilayer percep⁃trons,and random forests.Consequently,the correlation between traffic states and traffic conflicts is es⁃tablished.The study shows that the traffic state variables used in the prediction method significantly con⁃tribute to the probability of traffic conflicts.The random forest model can effectively predict the risk of traf⁃fic conflicts based on the traffic state parameters collected in 20 seconds,with a prediction accuracy of up to 97%.
作者
贾磊
李清勇
俞浩敏
JIA Lei;LI Qingyong;YU Haomin(Beijing Key Lab of Traffic Data Analysis and Mining,Beijing Jiaotong University,Beijing 100044,China;Frontiers Science Center for Smart High-speed Railway System,Beijing Jiaotong University,Beijing 100044,China)
出处
《北京交通大学学报》
CAS
CSCD
北大核心
2023年第3期61-69,共9页
JOURNAL OF BEIJING JIAOTONG UNIVERSITY
基金
中央高校基本科研业务费专项资金(2022JBQY007)
国家自然科学基金(62276019)。
关键词
交通安全
碰撞时间
交通冲突预测
隧道风险
机器学习
traffic safety
time-to-collision
traffic conflict prediction
tunnel risk
machine learning