【作者】 王鑫；

【导师】 郭应时；

【作者基本信息】 长安大学 ， 车辆工程（专业学位）， 2020， 硕士

【摘要】 在“人-车-路”的闭环道路交通系统中,人是最不稳定的一环,人的分心行为会给交通安全带来极大的隐患。相较于其他分心干扰方式,驾驶时使用手机带来的危害更高。将使用手机的分心行为和常见的跟车场景相结合,建立分心驾驶时,不同工作负荷的跟车碰撞风险辨识模型,有助于提高道路交通安全,减少事故发生。基于对已有研究的分析总结,本文设计了微信文字、微信语音、手持、免提、无次任务五种驾驶任务。利用DLab驾驶人因记录分析系统,开展模拟驾驶试验,采集驾驶人视觉行为、车辆运动状态、生理指标三方面数据,研究不同次任务下的驾驶特征及各任务之间的差异性,最终筛选出注视区域信息熵、有效注视时长、垂直扫视幅度、眨眼频率、车速标准差、横向位置标准差、横向加速度标准差、心率增长率、RR间期9个指标作为确定工作负荷的基本参数,利用主成分分析和K-均值聚类确定出驾驶人的分心工作负荷。基于不同的工作负荷,利用网格优化SVM算法,建立了驾驶人的跟车风险辨识模型,并加以验证。主要研究结论如下:视觉上,不论驾驶人是否处于分心状态,85%以上的视觉行为为注视,跟车时驾驶人的注视点主要集中于视野正前方前车的位置,位于水平(0°,10°],垂直(-10°,0°]。在执行微信文字和语音时,驾驶人视觉指标中注视区域信息熵、扫视速度、频率、幅度、持时增加,有效注视时长缩短。而微信语音次任务涉及视觉分心时间短,各项指标较微信文字变化小。执行手持和免提任务时,驾驶人以认知分心为主,扫视行为指标差异均不显著,由于手持通话涉及到驾驶人的操作分心,注意力更难集中,注视区域信息熵较高,有效注视时长较低,眨眼频率较高。车辆运动状态上,分心会导致驾驶人对车辆的控制变差。以视觉分心为主的微信文字和语音次任务,会导致车速标准差、横向位置标准差、横向加速度标准差显著增加,以认知分心为主的手持和免提次任务,对车辆控制影响相对小,与无次任务相比各项指标均不显著。生理上,由于心率存在一定的滞后性,分心驾驶时各指标变化不显著。基于生理测量的方式,采用驾驶人的视觉、车辆的运动状态、生理指标可以用于确定分心驾驶的工作负荷,结合驾驶人的反应时间验证,结果一致。针对驾驶人的不同工作负荷,采用网格优化SVM算法,对跟车风险状态有很好的预测结果,准确率为90.5%。对不同时间点的研究结果表明,在事故发生前1秒的预测效果最好,准确率91.3%、真正率92.3%、真负率为90.9%。更多还原

【Abstract】 In the driver-vehicle-road closed-loop road transportation system,the driver is the most unstable link.Driving distraction brings great hidden dangers to traffic safety.Compared with other methods of distraction interference,Distractions are more harmful by mobile phones.Combining the distraction behavior of using mobile phones with common car-following scenarios,establishing a recognition models of collision risk of following vehicles with different workloads during distracted driving can help improve road traffic safety and reduce accidents.Based on the analysis and summary of the existing research,five tasks(We Chat text,We Chat speech,handheld,hands-free,and normal driving)were designed in the paper.Based on the DLab driver’s factor analysis system,conduct simulated driving tests,collect data on visual behavior,vehicle motion,and physiological indicators,study the driving characteristics under different tasks and the differences between the tasks,and finally screen Nine indicators of the entropy of distribution of the fixation point,effective gaze duration,vertical saccade amplitude,blink frequency,standard deviation of vehicle speed,standard deviation of lateral position,standard deviation of lateral acceleration,heart rate growth rate,and RR interval were used as the basic parameters for determining the workload.Principal component analysis and K-means clustering were used to determine the workload of the driver.Based on different workloads,the grid-optimized SVM algorithm was used to establish a driver’s follow-up warning model and verified it.The main research conclusions are as follows:Visually,no matter whether the driver is distracted or not,more than 85% of the visual behavior is gaze.The driver ’s gaze point when following the car is mainly concentrated on the position of the car in front of the field of view,located at the level(0°,10°],vertical(-10°,0°].When performing We Chat text and speech,the driver ’s visual indicators of the entropy of distribution of the fixation point,saccade speed,frequency,amplitude,and duration increase,and the effective gaze duration shortens.The visual distraction time is short in the subtask of We Chat speech,and the indicators change less than We Chat text.When performing hand-held and hands-free tasks,the driver is mainly cognitively distracted,and the saccade behavior indicators are not significantly different.Because the hand-held call involves the driver’s operation distraction makes it harder to concentrate,the information entropy in the gaze area is higher,the effective gaze duration is lower,and the blink frequency is higher.In the vehicle motion state,distraction will cause the driver’s control of the vehicle to deteriorate.We Chat text and voice subtasks that mainly focus on visual distraction will result in a significant increase in the standard deviation of vehicle speed,horizontal position,and lateral acceleration.Handheld and hands-free subtasks with cognitive distraction as the main task have relatively little impact on vehicle control,and the indicators are not significant compared to those without subtasks.Physiologically,due to a certain hysteresis in heart rate,the changes in various indicators during distracted driving are not significant.Based on physiological measurements,the driver’s vision,vehicle motion status,and physiological indicators can be used to determine the workload of distracted driving.Combined with the driver’s response time verification,the results are consistent.Aiming at the different workloads of drivers,the grid-optimized SVM algorithm is adopted,which has a good prediction result for the risk status of following car,with an accuracy rate of 90.5%.Research results at different time points show that the prediction effect 1 second before the accident occurs is the best,with an accuracy rate of 91.3%,a true rate of 92.3%,and a true negative rate of 90.9%.更多还原