Driver Behavior Identification with Facial Recognition and Deep Learning

Driving entails a series of actions demanding intense focus. Some- times, these actions get overshadowed by other behaviors such as enjoy engaging in ingesting activities, conversing, using a cellphone, tuning the radio, or even drowsiness. These behaviors stand as the primary culprits behind today's traffic accidents. Thus, it's imperative to create applications that can forewarn drivers. This study proposes a streamlined convolutional neural network structure for identifying driver behaviors, thereby enhancing the accuracy of the alert system and minimizing traffic collisions. This network combines modules for extracting features and making classifications. By merging these components, the feature extraction within the neural network is designed to be more potent and efficient. The model aims to decrease computational costs and amplify data flow within the network by employing depth-wise separable convolutions. Additionally, the Convolution Block Attention Module aids the network in prioritizing vital features, resulting in improved performance for various computer vision tasks. The classifier unit includes global average pooling and a softmax layer to compute class probabilities. The overall architecture optimizes network parameters while upholding classification precision. Three standard datasets are used for both training and assessing the whole network.