考虑驾驶人性别差异的绿灯倒计时时长推测  

Prediction of green light countdown duration considering driver gender differences

作　　者：赵荣达 夏玉兰 杨榆璋 刘兵 赵伟 刘晔 谢济铭 ZHAO Rongda;XIA Yulan;YANG Yuzhang;LIU Bing;ZHAO Wei;LIU Ye;XIE Jiming(Yunnan Communications Investment and Construction Group Co.,Ltd.,Kunming 650103,China;Faculty of Transportation Engineering,Kunming University of Science and Technology,Kunming 650504,China)

机构地区：[1]云南省交通投资建设集团有限公司,云南昆明650103 [2]昆明理工大学交通工程学院,云南昆明650504

出　　处：《广西大学学报（自然科学版）》2023年第1期90-102,共13页Journal of Guangxi University（Natural Science Edition）

基　　金：国家自然科学基金项目(71861016);云南交投集团科技研发项目(YCIC-YF-2021-05)。

摘　　要：为了提升高级驾驶辅助系统(advanced driving assistance system, ADAS)在绿灯倒计时长(green light countdown time, GSCT)类型的辅助驾驶决策性,首先,通过实车实验采集到信号交叉口不同绿灯倒计时时长下的1 475 524条自然驾驶行为数据,包括18名驾驶员在不同绿灯倒计时时长下,分别通过城市道路18个信号交叉口前150 m情境的驾驶行为数据;然后,分析驾驶决策、速度、平均瞳孔大小位置、心电、肌电在男、女性驾驶人之间的差异,得到男、女性驾驶人在绿灯20 s倒计时期间的注视特性平稳、皮电特性最为稳定,在绿灯9 s倒计时期间男、女驾驶人心理特性最为稳定的结论;最后,利用KNN、SVM、GBDT等3种经典机器学习方法,基于前述5类特征参数,建立考虑多特征变量的信号倒计时长判断模型。结果表明:集成学习GBDT模型判断准确率为88.1%,精确率均值为85.4%,AUC均值为0.98,有助于ADAS提供决策支持和理论支撑,可为不同性别驾驶员在城市道路信号交叉口前提供速度调节辅助信息。In order to improve the decision making of the ADAS(advanced driving assistance system)in the GSCT(green light countdown time)type of assisted driving.Firstly,1475524 natural driving behaviours were collected from different green light countdown times at signal intersections,including the driving behaviours of 18 drivers who drove through the first 150 m of 18 signal intersections on urban roads at different green light countdown times.Then,the differences in driving decision,speed,average pupil size position,ECG and EMG between male and female drivers were analyzed,and the conclusion that the gaze characteristics of male and female drivers were smooth and the electrodermal characteristics were the most stable during the 20 s countdown to the green light,and the mental characteristics of male and female drivers were the most stable during the 9 s countdown to the green light;finally,using KNN,SVM,GBDT three types of classical machine learning methods,based on the aforementioned five types of feature parameters,to establish a signal countdown long judgment model considering multiple feature variables.The results show that the integrated learning GBDT model has an accuracy of 88.1%,an accuracy of 85.4%and an AUC of 0.98,which can help ADAS to provide decision support and theoretical support,and can provide speed regulation assistance information for drivers of different genders before signal intersections on urban roads.

关 键 词：交通工程 绿灯倒计时 驾驶特性 预测模型 

分 类 号：U491[交通运输工程—交通运输规划与管理]