面向闯红灯警告的驾驶员停驶行为判别研究

Predicting Individual Driver Stop-and-go Behavior to Intersection Red Light Running Warning System

面向道路交叉口闯红灯警告(RLRW),提出了通过进口道车辆运行参数,预测驾驶员将在停止线前停驶还是采取通过行为的判别方法。首先综述了闯红灯警告系统的产生与发展现状,采用二项Logistic回归方法分析数据之间的相关性并提出了一种新的驾驶员停驶行为预判模型,用城市道路实测数据进行了参数估计及检验,引入了描述系统性能的ROC曲线并将其用于模型性能评价。本研究采用的数据采集自上海城市道路交叉口进口道,通过Autoscope视频检测系统提取特征数据。研究表明,当前速度与上游30 m处速度差或速度比最适合作为解释变量预测驾驶员停驶行为,而非以往研究中采用的速度、加速度。距离停车线40 m作为计算点可获得较好的预测效果。

An approach to predict vehicle stop-or-go intention at intersection approaches using vehicle kinematical parameters to red light running warning(RLRW) was proposed.Current situation and development of red light running warning was reviewed first.An innovative binary logistic regression model was put forward to analyze data dependence and predict driver stop-or-go behavior.Parameters were estimated and tested by data from urban road.ROC curve was introduced in model evaluation.Data was collected from urban road intersection approaches in Shanghai and processed by Autoscope video detection system.Research result shows that velocity deference or velocity ratio between detect spot and the spot of 30 m upstream is the parameter which has the strongest correlation to stop-or-go behavior instead of velocity and acceleration.Good prediction results can be got if 40 m from stop line is taken as calculation location.