Driver Facial Emotion Monitoring Method Based on Deep Learning

With the development of the automotive industry, vehicle driving safety has become an important research topic. The inability of drivers to control their emotions is considered a key factor affecting driving safety. To prevent the negative impact of drivers’ negative emotions on driving behavior and road safety, this paper proposes a deep learning based facial emotion monitoring method for drivers, which monitors their emotions in real-time. The DSCNN-ECA network model method in this article takes the improved lightweight VGG network as the basic skeleton, introduces deep separable convolution method, and adds ECA attention mechanism. The method in this article adds ReLu activation function at the network layer to extract more representative network features. In addition, adding Dropout makes the network less prone to overfitting. The experimental verification showed that the method proposed in this paper was trained and tested on the Fer2013 and RAFDB datasets, respectively, and its facial expression recognition accuracy reached 72.14% and 88.26%, respectively. In addition, the experiment also verified that the improved neural network in this paper has higher accuracy and robustness than traditional convolutional neural networks.