【摘要】 对驾驶员在驾驶过程中的疲劳状态进行实时准确的预判，可减少由于驾驶疲劳引发的交通事故。首先，通过无线体域网采集了12名驾驶员在真实驾驶过程中的多模态特征，提取了脑电、肌电、呼吸等3种生理信号的特征参数近似熵，其中基于畸变能密度理论(DED)确定肌电信号的采集位置为颈6棘突旁开2 cm处的上斜方肌；然后，通过模糊C聚类方法分析了3种特征参数组合对疲劳状态的反映效果；最后，建立基于马氏距离理论的真实驾驶疲劳判别模型。研究结果表明，驾驶员颈6部位比颈7部位肌电信号的ApEn值显著下降(P<0.05),表明颈6处肌肉比颈7处肌肉对驾驶员的疲劳状态反映更为敏感，实际检测结果与畸变能密度理论计算结果一致，证明了该位置提取肌电信号的正确性和准确性；脑、肌、呼吸这3种生理信号的ApEn值均随驾驶时间的延长呈递减变化，驾驶约90 min时递减趋势变缓，表明驾驶员进入疲劳状态；通过模糊C聚类分析可知，当脑电与肌电ApEn组合时，清醒与疲劳的概率分布界限清晰，可有效反映驾驶疲劳状态；以脑电和肌电近似熵为自变量，基于马氏距离理论建立真实驾驶过程的疲劳判别模型，其测试集准确率达90.92%,表明该模型能够比较准确的判别出驾驶员的疲劳状态。更多还原

【Abstract】 In order to discriminate driver fatigue accurately in real-time and reduce the traffic accidents caused by driver fatigue, physiological signals of 12 subjects in actual driving were recorded by wireless body area network（WBAN）, and approximate entropy（ApEn） of electroencephalograph（EEG）, electromyography（EMG） and respiration（RESP） signals were extracted. The upper trapeziuses at 2 cm of both sides of 6^th spinous process were determined as the data acquisition positions of EMG based on distortion energy density（DED） theory. Then the discriminant degree of their combination was analyzed by the fuzzy C-clustering method. Finally, a discriminant model on driver fatigue was built based on Mahalanobis distance theory. The experimental results showed that the decreasing trend of the upper trapezius at 6^th spinous process was more obvious than that at 7^th spinous process, and the significant index P<0.05, indicating the muscles at 6^th spinous process were more sensitive for driver fatigue. The actual testing result was consistent with the calculation result of DED theory, and verified the correctness of acquisition position of EMG. During the actual driving, the ApEns of EEG, EMG and RESP signals decreased. After about 90 min, the decreasing trend slowed down, indicating the deeper fatigue. By the fuzzy C-clustering analysis, in the case of the combination of EEG-EMG, obvious discrimination of the probability distribution between normal and fatigued state were detected, and they were selected as independent variables. Finally, a discriminant model on driver fatigue based on Mahalanobis distance theory was built, and its accuracy was up to 90.92%, which effectively discriminated the driver fatigue.更多还原