模拟驾驶环境下驾驶人分心状态判别

为了探寻驾驶人分心判别方法，构建了驾驶人分心状态判别模型。首先设计分心模拟驾驶试验，采集正常驾驶和发送语音信息过程中的驾驶绩效特征和驾驶人眼动特征数据，建立驾驶人分心状态判别指标备选集；其次，采用基因选择算法对备选指标进行筛选，得到29个备选指标的重要度排序；然后，依次选取重要度较高的部分指标作为BP神经网络的输入指标，利用遗传算法（GA）全局搜索的性能优化BP神经网络的初始权值和阈值，将优化后的GA-BP神经网络作为弱分类器，再将多个弱分类器组合成Adaboost强分类器，建立基于Adaboost-GA-BP组合算法的驾驶人分心状态判别模型；最后，利用模拟驾驶器试验平台采集的数据计算不同判别指标数量下模型的性能，从而确定最优判别指标，并对模型进行验证和评价。结果表明：模型最优判别指标为重要度排序中前14个指标；模型能够准确识别驾驶人分心状态，判别精度为95.09%；与BP神经网络算法、GA-BP神经网络算法和Adaboost-BP神经网络算法相比，Adaboost-GA-BP组合算法在准确率、精准率、召回率、F₁值和ROC曲线等模型性能方面均最优。建立的模型能够有效判别驾驶人分心状态，可为驾驶人分心预警系统和分心控制策略提供依据。

Driver's Distraction States Identification in Simulating Driving Environment

To explore the recogonition method of driver distraction states, an identification model of driver distraction states was established. Therefore, an identification model to explore the recognition method of driver distraction states was established. First, this paper describes the design of a distracted-driver driving experiment conducted to collect experimental data, which includes the driving performance and driver's eye movement features during normal and distracted driving conditions, and the establishment of a subset of alternative indicators for driver distraction states. Second, support vector machine-recursive feature elimination (SVM-RFE) was used to rank these indicators, and the importance sequence of indicators was obtained. Third, some of the importance indicators were successively selected as the input of a back propagation (BP) neural network. Owing to its global search performance, the genetic algorithm was used for each BP neural network classification model for optimizing the weights and thresholds. The optimized BP neural network model was applied as a new weak classifier, and using the AdaBoost algorithm, many of these weak classifiers were composed into a strong classifier model. The Adaboost-GA-BP combined model was used for an identification of the driver distraction states. Finally, the performances of the models under different indicators were calculated using the experimental data to determine the optimal discriminant indicators, and the model was then validated and evaluated. The results show that the optimal features of the model are the 14 indicators of the importance sequence, and that the recognition accuracy of the driver distraction states is 95.09%. Compared to the BP, GA-BP, and Adaboost-BP models, the Adaboost-GA-BP model is superior to the other three methods in terms of accuracy, precision, recall, F₁-measure, and ROC curve. This model can effectively determine the driver distraction state, which can provide supporting data for a driver distraction warning system and a control strategy.