Vision Based Classification of Nocturnal Road Traffic Using a Custom Deep Convolution Neural Network

Abstract

Intelligent Transport Systems (ITS) have emerged as an efficient solution for enhancing road utilization efficiency, ensuring convenient and safe transportation, and reducing energy consumption. This research focuses on addressing the challenge of accurately estimating road traffic congestion, specifically in low visibility and adverse weather conditions such as rainy, overcast, and sunny weather. To tackle this issue, we propose a novel custom macroscopic approach for categorizing road traffic congestion using videos captured during nighttime. The proposed method leverages the power of deep convolutional neural networks (DCNN) to classify traffic into three distinct categories. A custom deep CNN model is developed and trained on nighttime videos from the UCSD public dataset, using 100 epochs. Through a rigorous evaluation by means of this dataset, our model achieves an impressive Correct Classification Rate (CCR) of 98.91%, surpassing the previously known state-of-the-art CCR of 89.47%. This exceptional precision in estimating road traffic congestion in challenging low visibility conditions is achieved thanks to the integration of advanced deep learning techniques and CNNs. The proposed method can be integrated as part of a traffic monitoring system, thus contributing to the advancement of Intelligent Transport Systems and their potential to create cleaner, safer, and more efficient transportation networks.