1.1 Importance of the drowsiness detecting
Driving has become one of the most indispensable cognitive behaviors in our daily life. Such a cognitive performance highly involves attention, decision making, information perception and awareness, and coordination of sensorimotor systems.
Therefore, decrease in driver’s attention or, more precisely, vigilance level may deteriorate driver’s driving performance and potentially cause car accidents. National Highway Traffic Safety Administration (NHTSA) in the US reported at least 100,000 car crashes were caused by drivers’ falling asleep [4]. Other studies also pointed out that fatigue, which, in turn, caused drowsiness or falling asleep, was one of the major causes of car accidents [1]-[3]. Such dramatic amount of tragic events caused by drivers’ drowsiness call for the attention and demand in devising an online, real-time drowsiness detection and monitoring system to assist and warn drivers when they become drowsy, such that we can best reduce the car accidents caused by drivers’
decreasing driving performance due to drowsiness.
1.2 Drowsiness measurements
In previous studies, drivers’ drowsiness states were mostly derived from different physiological measures based on either signal- or image-based technique.
The image-based technique employs video camera to detect, such as, eyelid closure, eyes’ gaze positions, head movements, etc., and derive, for example, the consecutive time periods of eyes closing, duration of gaze fixation, or alike, from such measures
to correlate with drivers’ drowsiness levels [7]-[13]. Such image-based technique requires nearly no preparation, in terms of applying electrodes or sensors, to drivers.
However, image-based methods may suffer from the environments with which the video cameras need to interact. For example, in the limited space of driving cabin, it is difficult to find a place to mount two video cameras, in the same time, without blocking the perception of the video cams by the handling wheel.
1.2.1 Signal-based drowsiness system
Most signal-based drowsiness monitoring systems use electrocardiograph (ECG), electroencephalograph (EEG) [17]-[22], or electrooculograph (EOG) [15][16]
to monitor drivers’ physiological changes related to their drowsiness levels. However, among these physiological changes, heart rate variability (HRV) [14] derived from ECG measures can be altered by all sorts of physiological or cognitive states and lack of specificity as an index of drowsiness levels. On the other hand, although EOG was used to index the decline of saccade frequency and velocity, which was proved highly related to the driving performance, it suffered from long average windows to establish the evidence for drowsiness and could not be used as a real-time warning system.
Other method, such as monitoring the patterns of drivers’ moving handle wheel as used by Toyota Motor Company[5][6], may also highly depend on drivers’ driving behaviors and experience, road conditions, and all other environmental variables, and thus is difficult to be generalized for regular use. As a result, EEG remains the most popular modality used to monitor drowsiness state in real-time.
1.2.2 A Better EEG Measure Technology
The power changes in EEG alpha (8-12 Hz) and theta (4-7 Hz) bands have been widely used to index the alertness levels of human subjects in other literature [23]. Recently, novel dry EEG electrodes based on Micro-Electro-Mechanical System (MEMS) technology have been invented and introduced to build wireless EEG acquisition and analysis systems [29]. This, in turn, greatly advanced the EEG recording technique in real-time operational environment and thus achieved EEG-based drowsiness monitoring system being used in driving simulation and even in real driving conditions. In the past, EEG has long been remained one of the laboratory devices for recording human brain waves due mainly to the professional preparation with injecting conducting paste into the electrodes to maintain the low contact impedance with scalp needed to assure good quality EEG signals. Let along the professional clean-up procedure needed after every experimental session. Such inconvenience prevent EEG device from being used in real operational environment, which needs easy-to-prepare and easy-to-maintain as well as wearable and long-term operation based only on battery power. The state-of-the-art dry MEMS EEG electrodes well fulfill the requirements as an easy-to-apply front end for EEG signal acquisition in the operational environment.
1.3 The frontal signals
However, as the current development of the dry MEMS electrodes, which consist of self-stabilized pin-shaped microelectrodes, they may not be easily applied to the hairy scalp surface. As a result, it may be more applicable to apply the dry MEMS electrode on the human forehead surface. Nonetheless, although the dry MEMS electrodes were already implemented on a baseball cap so that they can be
easily applied to human forehead surface with neither skin preparation (e.g., scratch the skin surface of scalp) nor application of conducting gel, it raised another issue here if the EEG signals acquired from forehead channels contain any detectable feature revealing drowsiness level.
1.4 The aims of this study
Therefore, here, we would like to explicitly compare the drowsiness features extracted using independent component analysis (ICA) to decompose the EEG data acquired simultaneously from both standard EEG channels as well as the forehead channels in a long-distance driving EEG experiment on a virtual-reality moving platform. In such comparisons, the drowsiness level-dependents components are extracted and selected by correlating the EEG spectra with the driving performance, which is used to index subjects’ drowsiness level as used in the previous studies [24][25]. In addition, the source of these drowsiness level-dependent components will be compared to see if these sources are commonly derived from the same EEG process, which may be located in the parietal/occipital cortex and mainly make up most of the drowsiness relevant EEG components as found in the past. The alternative can be that the drowsiness component extracted from the forehead EEG channels can be a useful feature, which is located in close to the frontal brain areas and can be readily picked by forehead EEG setup with novel dry MEMS EEG electrodes.