Improved Fatigue Detection Using Eye State Recognition with HOG-LBP

Abstract

Fatigue driving is one of the main factors in traffic accidents. Many driver fatigue alert systems have been developed to prevent fatal car accidents. Existing fatigue detection methods using image processing usually fail in face of illumination and occlusion variations. In this paper, we propose a novel fatigue detection method using eye state recognition with HOG-LBP fused features. Our method first proposes HOG-LBP fusion schemes to combine the advantages of HOG and LBP features. Taking concatenated or additive fused feature as input, we design deep neural networks and train the model on CEW eye dataset for eye state recognition. We give an in-depth analysis to select the optimal fused coefficient for the best model. Based on eye state prediction results, we extract two eye-related features and design a decision criterion for fatigue detection. Experimental results prove the feasibility of the proposed method in detecting drowsiness level at different driving conditions.