Mobile-based kernel-fuzzy-c-means-wavelet for driver fatigue prediction with cloud computing

A driver fatigue monitoring system with high precision could be a monetary countermeasure to reduce the road accidents. This study focuses on delivering fatigue prediction based on photoplethysmogram (PPG) and electrocardiogram (ECG) wavelet spectrum analysis. Specifically, an adaptive threshold method is utilized for PPG and ECG artifacts removal, peak and onset detection. Subsequently, the wavelet coefficients generated are further composed into very low frequency, low frequency and high frequency bands. Autonomous rule extraction is performed by using Kernel Fuzzy C-Means (Kernel FCM) with “if-then” rules to train the dataset for classifying driver vigilance level. By developing the hierarchical prediction model in smartphone device, it enabled the sensing data collection, fatigue level analysis, and warning sounded to driver when low arousal is detected, thus provide a safe and non-obstructive driving environment. Collected and analyzed data is uploaded to cloud server for remote monitoring. The experimental results validated prediction accuracy can be achieved at 96% to 98% on average across subjects.