無線前額腦波與眼動特徵擷取於增進駕駛疲勞警覺性

Abstract

隨著公路網絡的開發及完整性，開車所帶來的便利性日已漸增，但也伴隨著一定程度的風險，其中交通事故的發生率又以疲勞駕駛占了極大比例，最嚴重者可能導致死亡，若能藉由生理醫學電訊號得知駕駛人目前的生理狀態，評估駕駛行為下之即時反應時間，提醒駕駛人以提高警覺性，達到有效降低疲勞駕駛所導致的事故率。本研究探討駕駛狀態轉變下大腦及視覺神經網絡與其應用研究，以模擬駕車實驗場景誘發人們疲勞或瞌睡現象，找出與疲勞最相關之生理特徵組合，且與過去研究相互比較，並將本研究發現的生理特徵應用於可攜式無線人機介面系統。有鑒於過往文獻常以枕葉區（occipital lobe）為腦電訊號（EEG）量測位置，但在儀器的配置上可能會受到毛髮干擾而影響訊號之品質，因此本研究以前額葉腦區（frontal lobe）收錄腦電訊號，不僅可有效降低毛髮干擾，前額葉訊號也能同步擷取眨眼（eye-blink）特徵，故能實現以單一生理裝置達到多重生理訊號源之即時分析監測。在受測者方面，本研究收錄了十五位受測者，並以頻域分析及眨眼偵測作為系統建置之演算法工具，更同步使用線性回歸（linear regression）建置系統模型以估計駕駛行為下之即時反應時間。根據本系統之效能測試，在受測者內留ㄧ事件交叉驗證（With-in subject Leave-one-trial-out cross-validation）比較不同特徵組合之結果，發現以前額腦電及眨眼特徵之組合具有最佳的回歸效果，系統運算反應時間與真實駕駛反應時間之均方根誤差值（root mean square error）可降至0.034±0.019秒，確定係數值（coefficient of determination）可增加至0.885±0.057，並以數學統計檢定有無加入眨眼特徵之系統效能，會具有顯著差異性。在受測者內留ㄧ受測者交叉驗證（With-in subject Leave-one-subject-out cross-validation）之結果，對系統發出警報與非警報命中率亦有接近70%之準確性。本研究使用國立交通大學腦科學研究中心開發之無線可攜式腦波帽搭配即時線性運算，同步分析腦電及眼電訊號特徵，開發出一套可靠且精準的駕駛疲勞評估演算法，此演算法可以即時分析駕駛者的即時反應時間，再推算疲勞狀態的變化，進而評估駕駛者可能會進入疲勞打瞌睡的危險狀態，並將之與現代最普遍使用智慧型裝置整合成一套神經反饋系統，以實際道路駕駛即時監測，對人類的安全及生活更多加一分保障。

Driver fatigue problem is one of the important factors that cause traffic accidents. Recent years, many research had investigated that using EEG signals can effectively detect driver’s drowsiness level. However, real-time monitoring system is required to apply these fatigue level detection techniques in the practical application, especially in the real-road driving. In this study, we investigate to obtain a better and precise understanding of brain activities of mental fatigue under driving, which is of great benefit for devolvement of detection of driving fatigue system. The experiments based on a sustained-attention driving task, which was implemented in a virtual-reality (VR) driving simulator. We were performed for 15 subjects to study Electroencephalography (EEG) brain dynamics by using a mobile and wireless device (referred to herein as Mindo-4), which consists of foam-based electrodes, data acquisition module, Bluetooth transition module, and rechargeable battery. In addition, we also synchronous record Electrooculography (EOG) by using neuroscan system. In our preliminary results, we found the power spectral analysis showed that the dry EEG system could distinguish an alert EEG from a fatigue EEG by evaluating the spectral dynamics of delta and alpha activities. Furthermore, eye-blink analysis also showed that when people feel fatigue, the blink frequency will decrease, and the blink amplitude will increase. Based on the outstanding training results, the RMSE of system would decrease to 0.034±0.019 and the squared R would increase to 0.885±0.057 in the with-in subject leave-one-trial-out cross validation test. In statistics, the performances of system increased significantly after the features of eye-blink were added. As the system performance of with-in subject leave-one-subject-out cross validation test, the system accuracy would obtain 70% when multiple linear regression (MLR) model used combination features of the EEG power spectra of 1-30 Hz and eye-blink parameters. These results indicate that the combination of a smartphone and wireless EEG device constitutes an effective and easy wearable solution for detecting and preventing driver fatigue in real driving environments.