A novel driver fatigue assessment in uncertain traffic condition

In order to assess the driver fatigue in the dynamic, noisy and uncertain traffic conditions, this paper proposes a driver fatigue assessment system with a Bayesian network (BN). The multiple source feature data, such as percent eye closure and other behaviors that characterize a driver´s level of fatigue, sampled from driving subsystems, are processed into training and testing data sets. Using the training data, the assessment BN is modeled, and then testing features data sets presented to the assessment BN model to detect the onset of driver fatigue. By existing BN inference algorithms, and the inference result for driver fatigue assessment is provided. The presented approach achieves the assessment with not only complete evidences but also incomplete ones. Experimental results show that the proposed approach is more effective and robust in bringing out the driver fatigue classification than the traditional Radius basis function neural network method.