Development of a Method to Restore Driver Alertness

<p dir="ltr">Driver drowsiness has been a major concern for many years, as it considerably increases the risk of an accident. The Australian National Highway Traffic Safety Administration estimated that drowsy driving was responsible for 72,000 crashes, 44,000 injuries, and 800 deaths in 2013. Researchers are committed to investigating different methods to detect driver drowsiness and restore driver alertness to ameliorate this problem. Technologies such as Advanced Driver Assistance Systems are installed in modern vehicles to reduce the risk of driver drowsiness. </p><p dir="ltr">The present research aims to: 1) discover drowsiness detection methods suitable for drivers; and 2) develop a wearable alertness restoring system using a vibrating stimulus. The project uses a driving simulator to monitor the drowsiness levels of volunteer drivers (‘participants’). Subjective measurements utilised the Karolinska Sleepiness Scale to assess the drowsiness levels of the participants, whereas Heart Rate Variability (HRV) and Electroencephalogram (EEG) analysis were applied as objective measures of alertness. The frequency-domain features of the HRV data, which comprise the low frequency (0.04-0.15Hz) and high frequency (0.15-0.4Hz) components, were calculated from the Power Spectral Density (PSD). The experimental results show that the LF/HF ratios increased when a participant reported feeling drowsy. While HRV signals can reveal the different stages of alertness, their reliability has been challenged by some researchers. Therefore, EEG measurement was applied as another method of drowsiness detection. </p><p dir="ltr">EEG signals reflect the rhythmic activities of the brain cells which are highly correlated with different stages of alertness and sleep. EEG signals can be analysed in the following three ways: the wave feature description, the PSD, and Fractal Dimension (FD). The wave feature description is a time-domain analysis method, which interprets the EEG signals by directly analysing the EEG waveforms. On the other hand, PSD is a frequency-domain method, which analyses variations in the power of different frequency bands (α (8-13 Hz), β (13-30 Hz), θ (4-8 Hz), δ (0.5-4 Hz) and γ (30-50 Hz) waves) which characterise different stages of alertness. The FD method detects the alertness stage by describing the complexity of the EEG waveforms. FD analysis is not confounded by electrical noise from different sources. The researchers utilised an open-source EEG measurement platform (OpenBCI) to develop a portable EEG measurement instrument. Some customised components were applied to reduce the noise in EEG waveforms. </p><p dir="ltr">Vibration stimulation is a suitable approach for alertness restoration. To maximise the effectiveness of vibration stimulation, the principles of somatosensory sensations should be considered. Mechanoreceptors embedded in the skin are activated by particular frequencies and intensities of vibration. Thereafter the mechanoreceptors generate bioelectrical signals that are conveyed to the cerebral cortex through the spinal cord. It has been speculated that stimulation from some vibrational frequencies may entrain EEG brainwaves to oscillate at frequencies associated with greater alertness, thereby restoring alertness. Alternatively, the perception of unpredictable patterns of vibration may trigger attentional systems in the frontal lobe, thereby assisting in the maintenance of alertness and attention.</p><p dir="ltr"> A suitable actuator is an important requirement for developing a high-quality vibration stimulator. The following are the three kinds of actuators available: Eccentric Rotating Mass, Linear Resonant Actuator, and Piezoelectric. The main functional parameters of these actuators, such as response time, response frequency and maximum acceleration, were considered before the development of a prototype device that uses vibration to stimulate alertness. The researchers developed an actuator prototype, which includes a linear resonant actuator and a power amplifier on a printed circuit board. It can be driven by a computer using designed analogue signals. To test the effectiveness of prototypes, KSS, HRV and EEG were used to measure variations in participants’ alertness levels. </p><p dir="ltr">In the early stages of this research, LF/HF ratios were used to investigate the effect of whole-body vibration on driver alertness. The results demonstrated that low-frequency vibration increases drowsiness. HRV data supported this conclusion by demonstrating that LF/HF ratios in the vibration condition increased, whereas the LF/HF ratio displayed greater variability in the vibration condition when compared to the control condition. </p><p dir="ltr">In the later stages of this research, a wrist vibration actuator was used to stimulate alertness. The actuator used a specific pattern of vibration, which had been designed to reverse drowsiness. However, the HRV results did not match the subjective results to a great extent and this method showed some limitations. First, the heart rate chest strap exhibited low recording accuracy. Second, the LF/HF ratio proved to be unstable for short-term HRV analysis. </p><p dir="ltr">Due to the unreliability of HRV analysis, EEG was applied as another objective measure of alertness. Furthermore, KSS and EEG supported the effectiveness of the actuator prototype. The KSS results showed that the subjective alertness levels of the participants significantly improved after the actuator was applied. Simultaneously, the α:β ratios of the brainwaves decreased and the FD values of the EEG waveforms increased significantly, thereby indicating that the prototype had a positive impact on restoring alertness. </p><p dir="ltr">Image-based Deep learning (DL) models were also applied for drowsiness detection in this research as a non-contact drowsiness detection method. The researchers developed three DL models using the video clips of the participants, which were recorded in different positions. The DL models can automatically extract the features behind the data and record these features for classification or regression purposes. The highest accuracy of the DL models in this research crossed 97 per cent. </p><p dir="ltr">The outcomes of this research can be applied in the form of an independent intelligent device incorporated into smartwatches (such as Apple Watch) to help drivers maintain alertness. DL models have the potential to be incorporated into the cabin of vehicles, monitor driver alertness and provide a warning when the driver reaches a predetermined level of drowsiness.</p>