基于操作信号和眼动信号的驾驶员技能等级评价模型

Evaluation About Driving Skill Level Base on Eye Movement and Operating Signal

利用驾驶员在环仿真实验平台,搭建了较为真实的交通场景,采集驾驶员眼动信号和操作信号,利用眼动信号关于感兴趣区域(车内区域、车辆前方近处区域、远方区域)注视时间占比的结果给驾驶员贴标签,以减少标签错误的可能性;采用支持向量机(SVM)方法,比较研究基于单一操作信号(转向盘转角中心、速度、侧向加速度)和多操作信号组合评价驾驶员的技能等级。结果表明:采用纵向速度和侧向加速度信号组合比单一操作信号以及其他操作信号的组合对驾驶员技能等级分类正确率高。同时,通过比较多种驾驶员技能等级评价模型,得出最佳驾驶员技能等级评价的经验模型0.3*exp(6*V)+0.7*(8*Ac),其分类正确率达到了90%。

In order to make the driver’s skill evaluation agree with the real situation,the more realistic scenarios are built in the driver-in-the-loop experiment platform in this paper,where the eye movement information and the operating signals are collected. The results of the focused time proportion on the interesting areas(e.g. the vehicle interior area,the near area to the vehicle front area,or the distant area)based on the eye movement information are used to label drivers,and it could decrease labeling error. Simultaneously,the accuracyof the single signal(speed,steering wheel center,lateral acceleration)and the combined operation signalare compared by SVM model. The results show that the combination of the longitudinal velocity and lateral acceleration achieves the highest accuracy. By comparing a variety of driver’s skill evaluationmodel based on different characteristics,it concludes that the driver skill evaluation empirical model is(0.3*exp(6*V)+0.7*(8*Ac)),and the classification accuracy could reach 90%.