Fatigue state recognition based on improved ICA-HMM

Many traffic accidents are caused by drivers with fatigue states. It is important to detect fatigue states of drivers from their eye opening degrees, namely normal, dozing and fatigue. The paper proposes a fatigue state recognition algorithm that combines independent component analysis (ICA) and one-dimensional hidden Markov model (HMM) together. The algorithm firstly does binarisation processing for colour images, and ICA algorithm is then used to extract fatigue states. FastICA algorithm is deployed to accelerate the speed of feature extraction, and one-dimensional HMM is finally used to recognise the eye fatigue degree for the fatigue state. The experiment results show that the algorithm can rapidly and effectively recognise the different fatigue states at the eyes area of the driver.