A Novel Machine Vision-Based 3D Facial Action Unit Identification for Fatigue Detection

Fatigue has been attributed to traffic accident with higher fatality rate and causes severe damage to the surroundings compared to accidents where drivers are alert. This study presents and demonstrates an innovative driver fatigue detection method based on fatigue related facial action units' (AU) identification employing a photometric stereo (PS) testbed for 3D AU reconstruction. Initially, normal vectors were extracted for 3D AUs, subsequently, 'Quiver/Bump map' were constructed from the normal vectors. The quiver maps were further utilized for training deep neural networks. The findings exhibit that the proposed method outperforms 2D image based classification in terms of validation accuracy, for AU1, AU15 and AU41 detection. A novel method incorporating machine vision for intelligent transportation has been developed and demonstrated for driver's fatigue detection. The proposed method is also compared to other established methods and significant (95%) improvement in terms of accuracy is achieved.