Robust Detection of Fatigue Parameters Based on Infrared Information

Driver fatigue is one of the major causes of traffic accidents, and this need has increased the amount of driver fatigue detection systems in vehicles in order to reduce human and material losses. This work puts forward an approach based on capturing near-infrared videos from a camera mounted inside the vehicle. Then, from the captured images and using image-processing techniques the eyes are detected. Next, features are extracted from eye images using several transforms and finally, the system detects if there is fatigue or not using a SVM as classifier. Throughout the recording, eye position is tracked with a low computational time and fatigue is analysed based on the percentage of eyelid closure. This approach has been developed on two public datasets. Our experiments were able to reach an eye recognition rate of up to 96.87% and our results showed that SVM with RBF kernel were 99.66% accurate on one of the databases used for the system training. This approach shows promising results in comparison with the state of the art and deep learning approaches in order to be implemented in real conditions.