城市快速路出口分流区犹豫行为识别研究

Research on the Identification of Indecisiveness Driving Behavior in the Exit Diversion Area of Urban Expressway

为了探明城市快速路出口分流区的车辆犹豫行为特性,针对犹豫车辆的特征及其识别方法进行研究。采用无人机在230 m的高空对出口分流区进行拍摄,获取车辆运行视频数据,并利用Deepsort-Yolov3多目标追踪算法,提取犹豫车辆及周边车辆的行驶轨迹数据。根据车辆轨迹特征的差异,将其分为变道型及迟缓型两类。通过对滞后系数、距出口距离、最大瞬时速度差等特征指标的分析,得出:两类犹豫车辆的滞后系数均集中在20%~30%,显著大于正常车辆;变道型犹豫车辆距出口的距离集中在-10~70 m,显著小于正常变道车辆的100~130 m;迟缓型犹豫车辆的最大瞬时速度差集中在4~7 m·s^(-1),显著大于正常直行车辆的0~2 m·s^(-1)。说明与正常行驶车辆相比,犹豫车辆在出口分流区减速幅度更大,变道型犹豫车辆变道时,距离出口过近,不利于后方车辆的跟驰,增加了出口区的冲突及安全风险。采用滞后系数与距出口距离作为变道型犹豫车辆的识别指标,采用滞后系数与最大瞬时速度差作为迟缓型犹豫车辆的识别指标,分别构建两类犹豫行为的样本集,并随机划分为训练集和测试集。根据训练集数据,基于支持向量机(SVM)对两类犹豫行为进行识别模型的建立,并利用测试集数据进行评估。测试集评估结果表明:基于SVM的变道型犹豫行为识别模型预测准确率为96.7%,召回率为93.3%,AUC为0.990;迟缓型犹豫行为识别模型预测准确率为88.2%,召回率为100%,AUC为1.000,均效果良好。研究成果可以为犹豫型驾驶行为的快速识别及分流区交通安全风险防控提供理论依据。

In order to find out the indecisiveness behavior characteristics of vehicles in urban expressway exit diversion areas,the features of vehicles and their identification methods were studied.Using unmanned aerial vehicle(UAV)to shoot the exit diversion areas at an altitude of 230mand obtaining the video of vehicle running,extracting the driving trajectory data of indecisiveness vehicles and their surrounding vehicles by the Deepsort-Yolov3multiple-object tracking algorithm,and dividing the indecisiveness vehicles into two types:lane-changing type and retardation type according to the difference of vehicle trajectory features.Through the analysis of characteristic index such as lag coefficient,distance to exit,and maximum instantaneous speed difference,etc.,it is concluded that the lag coefficients of the two types of indecisiveness vehicles are concentrated between 20%to 30%,which is significantly larger than that of normal vehicles.The distance to exit of the lane-changing indecisiveness vehicles ranges from-10to 70m,which is significantly smaller than 100~130mof normal lane-changing vehicles.The maximum instantaneous speed difference of the retardation indecisiveness vehicles is concentrated in 4~7m·s^(-1),which is significantly larger than 0~2m·s^(-1) of normal straight vehicles.It shows that compared with normal driving vehicles,the indecisiveness vehicle decelerates more in the exit diversion area,and the lane-changing indecisiveness vehicle is too close to the diversion line when changing lanes,which is not conducive to the following vehicles and increases the conflict and risk in the exit area.Using Lag coefficient and distance to exit as the identification indexes of the lane-changing indecisiveness vehicles,and lag coefficient and maximum instantaneous speed difference as the identification indexes of the retardation indecisiveness vehicles,sample sets of the two types of indecisiveness behaviors were respectively constructed and randomly divided into training set and test set.According to the data of the training set,the identification model of the two types of indecisiveness behaviors was established based on the support vector machine(SVM),and the test set data was used for evaluation.The test set evaluation results show that the prediction accuracy rate of the lane-changing indecisiveness behavior identification model based on SVM is 96.7%,the recall rate is 93.3%,and the AUC is 0.990.Prediction accuracy rate of the retardation indecisiveness behavior identification model is 88.2%,the recall rate is 100%,and the AUC is 1.000.The results are all good.The research results can provide a theoretical basis for the rapid identification of indecisiveness driving behavior and the prevention and control of traffic safety risks in diversion areas.