A Comprehensive Deep Learning Framework For Real Time Traffic Density Estimation And Distracted Driver Behavior Detection

Abstract

This study presents a thorough deep learning framework for detecting distracted driving behavior and estimating traffic density in real time, two essential elements of intelligent transportation systems. Two separately trained modules are included in the suggested method. The first enables effective monitoring of traffic congestion and road usage patterns by identifying and counting vehicles from traffic scene photographs using a YOLOv8 object detection model. The second module classifies driver behaviors into ten categories, such as safe driving, texting, phone use, and other frequent distractions, by analyzing in-car photos using a bespoke Convolutional Neural Network (CNN). To improve the resilience and generalization of the model, extensive preprocessing methods and data augmentation were used. The framework is compatible with automated visual data processing, works in real-time, and may be implemented in driver safety and traffic surveillance systems. This dual-model architecture uses sophisticated vision-based monitoring to help create safer and smarter roads. When traffic congestion and queue clearing are present, the density of traffic is extremely nonlinear. Complex nonlinearities cannot be handled by closed-mathematical versions of standard density estimation techniques, which makes room for data-driven strategies like machine learning techniques. Deep learning algorithms, which identify nonlinear and highly situation-dependent patterns, perform best in data-rich environments.