摘要
为比较基于Logistic回归的驾驶人危险认知模型(RPL模型)与避撞时间模型(TTC模型)对跟车状态的辨识能力,利用图像式汽车行驶记录仪采集城市交通环境下的驾驶行为数据,分析驾驶人制动行为与跟车状态判断之间的关系。基于信号检测论建立了RPL模型和TTC模型的性能评价方法,对2种模型的判别准确率、似然比和辨别力指数进行比较和分析。结果表明:RPL模型和TTC模型在城市低速近距离跟车工况下均具有较高的判别准确率;RPL模型的判别准确率达到93%,安全阈值为条件概率取0.4,TTC模型的判别准确率达到92%,安全阈值为5s;2种模型的似然比和辨别力指数接近,表明2种模型判别跟车危险程度的能力相近;与TTC模型相比,RPL模型具有更大的适用范围。
To compare the accuracy of driver risk perception model based on logistic regression（RPL model） and time-to-collision（TTC） model,driving data in actual traffic flow were collected by video based driving recorder and the relationship between driver braking behavior and the identification of following status was analyzed.A method of performance assessment for RPL and TTC models was proposed based on signal detection theory and the prediction accuracy,discriminability index and likelihood ratio of two models were compared and analyzed.The results show that both RPL and TTC models have satisfying accuracy in low-speed condition and short-distance car-following situation,RPL model reaches 93% with conditional probability of 0.4 as safety threshold and TTC model reaches 92% with time-to-collision of 5 s as safety threshold.The discriminability index and likelihood are similar for two models,so that two models have similar performance of distinguishing car-following risk,but compared with TTC model,RPL model has a slightly wider scope of application.
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2012年第6期123-128,共6页
China Journal of Highway and Transport
基金
国家高技术研究发展计划("八六三"计划)项目(2012AA112308
2011AA110402)
关键词
交通工程
驾驶人行为
信号检测论
危险认知
避撞时间
traffic engineering
driver behavior
signal detection theory
risk perception
time to collision
作者简介
林庆峰（1975-），男，吉林敦化人，讲师，工学博士，博士后，E—mail：1inqf@buaa．edu．cn。
