Yawning Detection Based on Gabor Wavelets and LDA

To improve driving safety,the authors propose an approach to locate a driver's mouth by a web camera and extract texture features from mouth comers for monitoring drivers' yawning.Firstly,it detects drivers' left and right mouth corners by gray projection based on the result of driver face detection,and then it extracts texture features of drivers' mouth corners by Gabor wavelets.Finally,LDA is used to classify Gabor features for yawning detection.The proposed approach is tested on 3 000 images from thirty subjects with variations in illuminations,poses,and facial accessories(glasses).Yawning is also detected by the ratio of mouth height to width as a baseline.Experiment results show that the proposed approach is suitable for real time yawning detection,Gabor features are more powerful than geometric features for yawning representation,and an average recognition rate of 91.97%is achieved which is much better than the baseline.