Deep-Learning-Based Microscopic Approach for Road Traffic Congestion Classification in Highway and Urban Roads Under Adverse Conditions

Abstract

Road traffic congestion is a significant issue in urban and highway roads since it can cause various adverse effects, such as increasing travel time, elevated levels of air pollution, and a higher likelihood of traffic accidents and collisions because of congested conditions and frustrated drivers. Road traffic state estimation systems may improve traffic fluidity and thus remedy these effects. We propose, in this paper, a new microscopic approach to categorize the traffic flow into three classes: light, medium, and heavy. The microscopic parameters are extracted using vehicle monitoring, which consists of vehicle detection, using the deep-learning You Only Look Once (version 5) algorithm, and vehicle tracking. The extracted parameters are input to a classifier. Three classifiers are tested: the support vector machine, K-nearest neighbor (KNN), and random forest classifiers. The performance of the proposed method is evaluated using videos from two datasets: the well-known University of California, San Diego dataset and the urban University at Albany Detection and Tracking dataset. With both datasets, we could achieve 100% classification accuracy. This highest score was obtained using only two extracted parameters and the simple KNN classifier, which reduces the computational load of the proposed method.