Car driver fatigue monitoring using Hidden Markov Models and Bayesian networks

In this paper, audio, heart rate, steering wheel, gas, clutch, and brake pedals positions are used to determine the level of fatigue for the driver. The system consists of three main modules. One decides the fatigue level based on the audio signal, another decides the fatigue level based on the heart rate and the other signals, and the last one combines the decisions from the first two modules using a Bayesian network. Hidden Markov Model (HMM) classifier is used to model the fatigue for both audio module and other signals module. The experimental results show that HMM performs better than Support Vector Machines (SVM) for both audio and other signals modules. The results also show that combining more than one decision improves the performance of the system.