Steering Towards Safety: Approach of CNN Architectures to Predict Distracted Driving

Distracted Driving is becoming the critical problem day by day causing accidents, fatalities and injuries worldwide. With the demand of automated intelligent transportation system, Artificial Intelligence (AI) with computer vision technologies will assure the driver safety with the help of automated monitoring system. The research addresses deep learning model implementation along with evaluation for detecting distracted driving incidents. The research investigates how CNN (Convolutional Neural Network) architectures including MobileNetV2, EfficientNet-B0, ResNet-152 and InceptionV3 function to detect driving behaviors using the State Farm Distracted Driver Detection Dataset with thousands of images. The models used are transfer learning methods along with image preprocessing to conduct their training that led to evaluations concerning classification accuracy, loss reduction and confusion matrix analysis. The MobileNetV2 and EfficientNet-B0 models outperformed with the 99.74% and 99.71% validation accuracies respectively. This makes it suitable for embedded system real-time deployments. The research shows how lightweight CNN models achieve successful performance in real-time driver monitoring applications which reduce distraction-caused accidents on the roads. Overall, the technology helps in detecting the distraction in the driving and warns the drivers about not paying the attention.