Development of a Digital Tool for Quantifying and Classifying Traffic Entities

Abstract Accelerated urbanization and rising traffic volumes require intelligent monitoring solutions. Technological evolution has enabled the development of software tools that were not possible in earlier periods. In our research a digital tool based on deep learning YOLO networks was used for detecting, classifying, and counting vehicles in real traffic. After analyzing existing methods, a custom dataset was created, images were labelled, and a YOLOv8 model was trained on five vehicle classes (cars, trucks, buses, trams). The model achieved an F1 score of 86% and a mAP above 90%, outperforming the pre-trained version. Detected vehicles were converted into standard traffic entities (VEs) to improve flow estimation accuracy. The results confirm the method’s feasibility and its potential for applications such as tolling, traffic analysis, and road maintenance. A set of model limitations was identified during the research, thus opening a new direction for further study. Future work will include the expanding of dataset and deploying the model for real-time monitoring, contributing to the digitalization of urban traffic management through AI-based tools.