Multi-Classifier for Highly Reliable Driver Drowsiness Detection in Android Platform

For the past decade, it is well defined in the literature that fatigue is one of the most prospective factor in affecting the driver behavior. This paper presents a novel evaluation of driver fatigue condition based on multi-classifier technique and fusion of attributes approach. The process involved fusion of attributes including image of eye movement and photoplethysmography (PPG) signals that are given as inputs to multi-classifier. In order to develop the best inference classifiers, artificial neural network (ANN), dynamic bayesian network (DBN), support vector machine (SVM), independent component analysis (ICA) and genetic algorithm (GA) were tested in our study. The output from each inference classifier are scaled and product in an intervention module to indicate driver aptitude in real-time. Implementation of monitoring system is practically designed in Android-based smartphone device where it can received all the sensory information from the dedicated sensors installed at the steering wheel via a small scale wireless sensor network. Device built-in front camera was utilized to capture driver facial image. No supplementary monitor is required to be installed in the vehicle as the all the information is to be displayed on the smartphone device itself. Warning system is triggered to warn driver once fatigue is suspected. System testing statistical results revealed that the manifold used of the proposed system demonstrates the advantages of performing information fusion, particularly with discrete methods, and the multi-classifier enabled a more authentic and ample driver fatigue evaluation.