Optimización de la gestión de semáforos en intersecciones urbanas mediante detección y clasificación de vehículos con inteligencia artificial

Abstract

This thesis presents a methodology for optimizing traffic signal management at urban intersections through the application of artificial intelligence based on convolutional neural networks. A vehicle detection and classification model were developed using the YOLOv8 architecture, trained on a custom dataset composed of images extracted from intersections in Cuenca, Ecuador. For object tracking and real-time vehicle counting, the DeepSORT algorithm was integrated, enabling accurate analysis of classified traffic flows. The system’s outputs were then incorporated into the SUMO traffic simulation platform to evaluate three scenarios: the current configuration with static signals, a modified layout with geometric adjustments, and an adaptive traffic control scheme informed by the detection model. Results demonstrate that an AI-based adaptive approach can substantially reduce waiting times without compromising service levels, thus improving operational efficiency in complex intersections. This framework, once refined, represents a scalable and technically viable solution for modern traffic management systems.