Desarrollo de un sistema inteligente de aforo vehicular en tiempo real mediante reconocimiento de imágenes y redes neuronales

Abstract

The acquisition of precise real-time vehicle flow data is essential for the management and optimization of transportation networks. Current methods in Bogotá, such as manual counting and fixed sensors, have significant limitations in coverage and accuracy. This project addresses these challenges by developing an intelligent real-time vehicle counting system using convolutional neural networks and computer vision. Utilizing the YOLOv8n model, vehicles were detected and classified into three categories: cars, motorcycles, and buses/trucks.

High-resolution videos were captured at a strategic point in Bogotá, and the images were processed to train the model. Data processing techniques included labeling, augmentation, and normalization of images. Evaluation metrics such as precision, recall, mAP, F1 Score, and confusion matrix analysis demonstrated the model's high effectiveness in vehicle detection. Additionally, processing speeds and system efficiency, crucial for real-time applications, were discussed.

The results show a high level of precision and recall, suggesting the system's feasibility for implementation in urban traffic management. This enhances traffic planning, reduces travel times and emissions, and provides reliable data for mobility decision-making.