Computer Vision-based Driver Fatigue Detection Framework with Personalization Threshold and Multi-feature Fusion

Abstract

This paper proposes a novel non-intrusive and computer vision-based framework for driver fatigue detection from the video. First, to improve the judging accuracyof the driver’s facial expressions, the personalized threshold is proposed insteadof the traditional average threshold. Secondly, in order to alleviate the impactof the lack of relevant public data on model training, the transfer learning isused to train the eye and mouth state classifier. Finally, to solve the problem oflow universality and accuracy of the driver fatigue detection caused caused byusing only one type of facial features, multiple features including appearance-based features and deep learning-based features are used to detect driver fatiguedriving. The experiment results indicate that our method achieves 92.21% F1score and 29 fps, and yields better performance than traditional methods on thepublic NTHU-DDD dataset.