CHAUFFEUR BEHAVIOR RECOGNITION USING FACE RECOGNITION AND DEEP LEARNING

Abstract

High levels of focus are necessary for safe driving, but these behaviors are frequently overridden by distractions like tiredness, eating, drinking, talking, and phone calls. Sadly, these distractions play a significant role in the worrying increase in traffic accidents nowadays. The creation of software that can proactively inform drivers is essential to resolving this pressing problem. This study suggests a novel, lightweight architecture for convolutional neural networks that is intended to recognize different driving styles, enabling warning systems to provide accurate information and dramatically lowering traffic collisions. This network effectively recognizes and categorizes driver behaviors by fusing feature extraction and classifier modules. By combining these elements, in the neural network, feature extraction is intended to be more effective and efficient. The model may lower computing costs and enhance information flow through the network by utilizing depth-wise separable convolutions and adaptive connections. Furthermore, the Convolution Block Attention Module can assist the network in prioritizing crucial features, improving performance across a range of computer vision applications. Using a global average pooling layer and a soft max layer, the classifier module successfully calculates class probabilities. This carefully thought-out architecture makes sure that the network parameterization is optimized while yet maintaining good classification accuracy. On three benchmark datasets, the entire network is painstakingly trained and carefully assessed, confirming its dependability and robustness.