网联与非网联环境下驾驶人换道意图识别研究

Driver Lane-changing Intention Recognition in Connected and Non-connected Environments

网联时代的到来必将改变驾驶人的注意力分配和环境感知能力,进而影响其认知与行为模式。此时,非网联环境中建立的换道意图模型是否继续适用于网联环境值得研究。因此,基于驾驶模拟器搭建了网联与非网联换道场景,对比分析了2种环境下驾驶人换道意图表征参数与换道意图识别模型。结果发现:网联环境下的换道意图时间窗口长度(6.6 s)比非网联环境(4.1 s)长了约60.98%。网联环境下,意图表征参数(车辆运动与驾驶人操作参数)波动幅度显著小于非网联环境。换道意图阶段,非网联环境下驾驶人的平均扫视速度、后视镜观看频次以及注视时间与网联环境下存在显著差异。模型对比发现,当提前0.5 s识别时,网联与非网联环境下的模型识别精度无明显差异,分别为98.25%和96.35%;当提前2 s识别时,网联环境下的模型识别精度(93.48%)显著高于非网联环境(85.68%);在提前3 s识别时,网联环境下的模型识别精度为92.23%,非网联环境下出现了训练不收敛的情况。综上可见,网联与非网联环境下驾驶人换道意图表征参数与模型识别精度存在较大的差异。此外,网联环境下驾驶人换道意图通过所提模型精确识别后利用网联通信方式发送至宏观交通系统或周围车辆,不仅有助于宏观交通系统对整个交通流的综合调控管理,还有助于周围车辆跟车目标的提前切换以及换道轨迹规划。

The arrival of the connected era will change drivers'attention span and environmental perception ability,affecting their cognitive and behavioral patterns.This raises the question of whether the lane-changing intention model established for non-connected environments will still apply to connected environments.In this study,connected and non-connected lane-changing scenarios were built based on a driving simulator,driver's lane-changing intention characterization parameters,and lane-changing intention recognition models for both environments.The results show that the time window length for lane-changing intention in a connected environment(6.6 s)is approximately 60.98%longer than that in a non-connected one(4.1 s).In a connected environment,the fluctuation range of the intention characterization parameters(vehicle motion and driver operation parameters)is significantly smaller than that in the non-connected environment.In the lane-changing intention stage,the mean saccade velocity,rearview mirror viewing frequency,and fixation time in the non-connected environment were all significantly different from those in the connected environment.The model comparison revealed that different recognition times produced different accuracy responses across both environments:①when the recognition is performed 0.5 s in advance,there is no significant difference between the connected(98.25%)and non-connected(96.35%)environments;②when it is performed 2 s in advance,there is a significantly higher recognition accuracy in the connected environment(93.48%)compared to that in the non-connected environment(85.68%);and③when it is performed 3 s in advance,the model recognition accuracy in the connected environment is 92.23%but does not converge in the non-connected environment.In summary,there is a significant difference between the lane-changing intention characterization parameters and the model recognition accuracy in both connected and non-connected environments.In addition,the model used in this study accurately identifies the driver's lane-changing intention in a connected environment,which is then sent to the macro-traffic system or surrounding vehicles via connected communication.The use of this model can thus help inform the macro-traffic system to comprehensively regulate and manage the entire traffic flow,as well as the advance switching of the ensuing targets of the surrounding vehicles and the planning of the lane-changing trajectories.