虛擬駕駛環境下腦波頻譜與反應時間之關聯

Abstract

疲勞駕駛不僅危險，並易造成交通事故，故瞌睡偵測系統的開發已成為駕駛安全上重要課題。本實驗讓受測者在虛擬實境（VR）下，進行事件相關車輛偏移（event-related lane-departure）實驗，實驗中同時量測受測者的腦電波（EEG）訊號，以了解受測者駕車行為反應與腦電波能量頻譜（power spectrum）之關聯。所錄得之腦電波訊號，在去除雜訊後，先以獨立成份分析（ICA）分出不同獨立訊號源，再將這些訊號源產生的腦波以時頻轉換（time-frequency transform）算出其頻譜。將所得頻譜依訊號源經過分群（clustering）並依相對應受測者反應時間排序後，觀察到兩側枕葉區（bilateral occipital）在alpha頻帶（band，頻率為8~12赫茲）上的腦波頻譜能量會隨反應時間增加而上升，但若反應時間更長，則其能量會下降；而在theta頻帶（頻率為4~7赫茲）的腦波頻譜能量則隨受測者反應時間增加持續而上升。實驗中亦觀察到若受測者反應時間增加，則該受測者通常會出現打瞌睡的行為。因前述腦區之腦波能量改變現象不論實驗中是否提供動態刺激均可觀察到，故該腦區之腦波特徵可用於設計瞌睡偵測器，以保障駕駛人安全。

Drowsy driving is a dangerous behavior and often results in a large number of fatal accidents each year; therefore, understanding the neural correlates of drowsy driving is crucial for the design and evaluation of devices for drowsiness detection. This study investigates the relation between spectral features of electroencephalo-graphic (EEG) signals and driving performance. Subjects participated in long-haul simulated driving experiments on a motion platform, during which driving trajectories and 30-channel EEG signals were recorded simultaneously. Driving performance was measured by reaction time (RT) as defined in an event-related lane-departure paradigm. Following artifact rejection on behavioral and EEG data, independent component analysis (ICA) was used to decompose EEG signals into independent brain processes, and power spectra were computed from the activation time course of each independent component. Independent components with similar features, including topographic maps, dipole sources, and power spectra, were then grouped into clusters across subjects. Across subjects, an independent component with sources in the bilateral oc-cipital regions showed prominent changes in EEG power spectra as reaction time to lane-departure events increased. The alpha-band (8-12 Hz) power increased as reaction time increased and started to decrease as reaction time further increased (> 3 sec); however, theta-band (4-7 Hz) power increased monotonically as reaction time increased. These spectral features were consistent in both motionless and motion conditions. Finally, the results of this study may provide useful information, such as the selection of optimal electrode locations and frequency bands, for the development of drowsiness detection devices.