A Transfer Learning Approach with Modified VGG 16 for Driving Behavior Detection in Intelligent Transportation Systems

The majority of driving errors resulting from poor driving habits, a lack of compliance with the law, and inadequate driving knowledge continue to be a major global problem for public safety. Delivering accurate and timely warnings to drivers for correction while also allowing law enforcement authorities to catch violators is a critical task. To address this problem, the paper examines the complex aspects of human behavior and the different variables that cause distractions on the road, and a highly reliable and robust transfer-learning-based AI-system has been proposed. Initially, a novel dataset with ten types of driving behaviors was developed. Afterward, a fine-tuned modified VGG16-model was applied to classify them. In addition, the model was trained by assessing the efficacy of four optimizers: Adam, SGD, AdaGrad, and RMSProp. The optimized modified VGG16 with the Adam Optimizer proved to be the best, attaining excellent accuracy, recall, F1 score, and precision, all of which were 99.86%. It illustrates the effectiveness of the proposed approach in significantly improving the accuracy of identifying distracted driver- behavior.