Classification and Prediction of Driver Emotional States Based on Facial Expressions in Intelligent Cockpits

With the rapid evolution of intelligent cockpit technology, effective classification and prediction of driver emotional states via facial expression recognition has emerged as a vital component for enhancing both safety and driving experience. This study introduces a comprehensive framework that leverages deep learning techniques to analyze driver facial expressions captured in real-time scenarios. In our approach, a convolutional neural network (CNN) based on the VGG16 architecture is employed to automatically extract meaningful features from facial images. Prior to model training, the dataset-sourced from a publicly available facial emotion recognition repository-is rigorously pre-processed through image enhancement, noise reduction, normalization, and resizing to ensure uniform quality and dimension. To overcome challenges posed by limited dataset size, data augmentation strategies such as random rotations, horizontal flipping, and brightness adjustments are applied, substantially expanding the training sample and enhancing model robustness. The refined CNN model is then trained to distinguish between key emotional states relevant to driving behavior, specifically focusing on happy, angry, and neutral expressions. Experimental results demonstrate that our model achieves high accuracy in classifying driver emotions, even under resource constraints, as evidenced by favorable loss and accuracy metrics over a limited number of epochs.