Weighted Multi-Task Vision Transformer for Distraction and Emotion Detection in Driving Safety

Detecting drivers' distraction and emotion has raised attention due to its importance in ensuring driving safety, especially with the increasing number of accidents caused by distracted or emotionally unstable drivers. Previous research has employed the multitasking method to detect these two factors simultaneously but paid insufficient attention to the emotion detection part. Meanwhile, existing publicly available datasets use side cameras to capture both distraction and emotion, which is impractical in real driving scenarios. To address these issues, in this paper we propose a vision transformer-based approach that enhances the emotion detection accuracy while maintaining the performance of distraction detection by balancing the two. Moreover, we generated a new dataset that includes a front view of the driver's face, which improves the accuracy of emotion detection. Evaluation results validate the effectiveness of the proposed approach and demonstrate the balanced importance of emotion detection in driving safety. Our proposed method presents valuable contributions towards enhancing the safety of driving by highlighting the significance of emotion detection and introducing practical solutions to improve its accuracy.