Vehicle Counting and Classification in Varying Traffic Conditions using Faster R-CNN

Abstract Traffic congestion is a persistent problem requiring efficient management solutions. A Traffic Management System (TMS) has been introduced to mitigate traffic congestion due to scientific and technological advancements. One of the primary problems in TMS is accurately identifying and quantifying moving vehicles, especially for complex data, such as one-way traffic flow videos. To address these problems, we propose enforcing the Faster R-CNN model. We systematize it in three main steps. The first step is data preprocessing, which converts the video into frames and improves the image quality. Then, we created a TMS with Faster R-CNN. Finally, the classification and counting of moving vehicles. There are five classes of vehicles: car, bus, truck, motorbike, and bicycle. Each vehicle was identified as having either moderate or heavy traffic flow. Based on the experimental results, we obtained the highest average counting and classification accuracy of 87.89% and 93.18%, respectively. Hence, our results can be used as a solid baseline for further TMS development and practical applications.