Nighttime Road Traffic Videos Classification based on a Custom Deep Convolutional Neural Network

Intelligent Transport Systems (ITS) focus on gathering, storing, and offering real-time road traffic data to enhance road efficiency, ensure safe and convenient transportation, and decrease energy usage. By integrating advanced technologies, ITS contributes to cleaner, safer, and more efficient transportation. Consequently, ITS has gained prominence in regulatory and legislative initiatives across nations. This research aims to develop a highly accurate algorithm for estimating road traffic density in challenging conditions, particularly low visibility, such as nighttime. In this study, we propose a novel macroscopic approach for categorizing road traffic congestion using videos recorded during nighttime. This method employs deep convolutional neural networks (DCNN) to classify traffic into three categories. We utilized a custom CNN model trained on nighttime videos from the UCSD public dataset (University of California San Diego) over 100 epochs. With this dataset, we achieved a Correct Classification Rate (CCR) of 98.91%, surpassing the known state-of-the-art CCR of 89.47%.