个体差异对转向指标疲劳辨识能力的影响分析

驾驶人个体差异是影响疲劳驾驶辨识准确性的重要因素。为了探究个体差异与基于转向行为的疲劳辨识效果之间的关系，量化个体差异对转向特征指标疲劳辨识能力的影响程度，通过自然驾驶试验，采集被试在清醒状态和疲劳状态下的真实驾驶行为数据，结合观察员问询打分和被试面部视频得到疲劳水平信息。设置双层滑动时间窗对每位驾驶人的自然驾驶行为数据进行处理，挖掘出9个疲劳驾驶转向特征指标。对每位驾驶人清醒和疲劳状态下的指标样本进行Wilcoxon检验，用Wilcoxon检验的|Z|值表示指标对驾驶疲劳的分类性能。以清醒和疲劳状态下指标有显著差异的被试数目最多为优化目标，得到指标最优的双层时间窗设定值。将|Z|值最大的被试逐个与其他被试两两组合，对清醒和疲劳状态下混合两被试指标样本数据进行Wilcoxon检验，得到被试组合指标的|Z|值。计算两被试的综合个体差异值，基于线性模型拟合两被试组合Wilcoxon检验的|Z|值和个体差异值，以拟合直线的斜率绝对值|k|量化个体差异对指标疲劳辨识能力的影响。研究得到基于自然驾驶行为数据的9个疲劳驾驶转向特征指标的最优时间窗，发现指标对疲劳驾驶的分类性能存在个体差异，并且指标的疲劳辨识能力会随个体差异增加而降低，进而影响基于转向行为指标疲劳辨识的准确性，其中方向盘转角下四分位标准差（X_q1std）的斜率绝对值最大（1.17），方向盘转角标准差（X_jstd）的斜率绝对值最小（0.44），疲劳辨识能力受个体差异影响最大和最小的指标分别是X_q1std和X_jstd。研究结果可为利用自然驾驶行为数据的疲劳驾驶特征提取及考虑个体差异的疲劳驾驶建模提供参考。

Analysis of Influence of Individual Differences on the Fatigue Identification Abilities of Steering Indicators

Individual differences among drivers are crucial factors affecting the accuracy of fatigue driving identification. This study aims to explore the relationship between individual differences and fatigue identification based on steering behavior and quantify the influence level of individual differences on the fatigue identification abilities of steering indicators. Through naturalistic driving experiments, naturalistic driving behavior data were collected in the awake and fatigued conditions, and the fatigue level information was obtained by combining inquiry-based scores obtained from observers and videos of facial expressions. The naturalistic driving data of each subject were processed using a double-layer sliding time window, and 9 characteristic indicators of fatigue driving were identified. For each subject, the indicator samples in the awake and fatigued statuses were analyzed via a Wilcoxon test, and the|Z|value of the Wilcoxon test was used to evaluate the classification performance of each indicator,with regard to driving fatigue. The number of subjects who exhibited significant differences between the awake and fatigued conditions was the optimization target, and the optimal parameters of the double-layer sliding time window were obtained for the indicators. The data obtained for the subject corresponding to the maximum|Z|value were individually combined with those obtained for each of the other subjects, and the combined indicator sample data of batches of two subjects were analyzed via the Wilcoxon test; thus, the indicator|Z|for the combined data was obtained. Then, the comprehensive individual difference values for the two subjects were calculated. The|Z|value for the combined data and the individual difference values were fitted based on a linear model, and the absolute value of the slope (|k|) of the fitted line was used to quantify the influence of individual differences on the fatigue identification effectiveness of the previously obtained indicators. Thus, the optimal double-layer sliding time windows of nine steering indicators for fatigue driving are obtained based on naturalistic driving behavior data. It is also found that there are individual differences in the classification performance of the indicators, with regard to fatigue driving; in addition, the fatigue identification performance of the indicators decreases with an increase in individual differences. Further, the accuracy of fatigue identification is affected depending on the steering behavior indicators employed.|k|of the standard deviation of steering wheel angle below lower quartile (X_q1std) is maximum (1.17) and that for standard deviation of steering wheel angle (X_jstd) is minimum (0.44), this indicates that the fatigue identification abilities affected by individual differences the most and least are X_q1std and X_jstd. The results can serve as a reference for the extraction of indicators of fatigue driving based on naturalistic driving behavior data and for the modeling of fatigue driving considering individual difference.