Thesis Medición de flujo vehicular y peatonal mediante reconocimiento de imágenes de la región Metropolitana de Santiago
Loading...
Date
2025-12
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Mecánica
Departament
Campus
Campus Santiago San Joaquín
Abstract
La gestión de la movilidad urbana y de la calidad del aire en la Región Metropolitana de Santiago requiere contar con información de flujo vehicular actualizada, confiable y de alta resolución temporal. Sin embargo, los métodos tradicionales de aforo presentan limitaciones en términos de costo, escalabilidad y capacidad de integración con sistemas ambientales, lo que dificulta la elaboración de inventarios de emisiones representativos. En este contexto, el presente trabajo de título, desarrollado en colaboración con el Laboratorio de Contaminación y Procesos Energéticos (LabCPEL) y el Gobierno Regional Metropolitano (GORE-RM), tiene como objetivo validar e implementar un sistema de monitoreo no intrusivo del tránsito vehicular basado en visión por computador e Internet de las Cosas (IoT), orientado a su aplicación en estudios de calidad del aire. La metodología consideró, en una primera etapa, la evaluación técnica y económica de distintas tecnologías de captura, seleccionándose una cámara industrial Milesight por sobre otras alternativas comerciales y prototipos locales, debido a su robustez, disponibilidad y capacidad de procesamiento en el borde (Edge Computing). Posteriormente, se implementaron y calibraron algoritmos de detección de objetos mediante redes neuronales convolucionales (YOLO) y técnicas de seguimiento (tracking), cuyo desempeño fue validado a través de campañas experimentales que compararon los conteos automáticos con registros manuales controlados. Este proceso permitió identificar y mitigar desafíos críticos asociados a la oclusión vehicular y al ángulo de incidencia del sistema de captura(...).
The management of urban mobility and air quality in the Santiago Metropolitan Region requires up-to-date, reliable, and high-temporal-resolution traffic flow information. However, traditional counting methods present limitations regarding cost, scalability, and integration capacity with environmental systems, hindering the development of representative emission inventories. In this context, this thesis work, developed in collaboration with the Laboratory of Contamination and Energy Processes (LabCPEL) and the Metropolitan Regional Government (GORE-RM), aims to validate and implement a non-intrusive vehicular traffic monitoring system based on computer vision and the Internet of Things (IoT), oriented towards its application in air quality studies. The methodology considered, in a first stage, the technical and economic evaluation of different capture technologies, selecting a Milesight industrial camera over other commercial alternatives and local prototypes due to its robustness, availability, and Edge Computing capabilities. Subsequently, object detection algorithms using convolutional neural networks (YOLO) and tracking techniques were implemented and calibrated. Their performance was validated through experimental campaigns comparing automatic counts with controlled manual records. This process allowed identifying and mitigating critical challenges associated with vehicular occlusion and the capture system's angle of incidence(...).
The management of urban mobility and air quality in the Santiago Metropolitan Region requires up-to-date, reliable, and high-temporal-resolution traffic flow information. However, traditional counting methods present limitations regarding cost, scalability, and integration capacity with environmental systems, hindering the development of representative emission inventories. In this context, this thesis work, developed in collaboration with the Laboratory of Contamination and Energy Processes (LabCPEL) and the Metropolitan Regional Government (GORE-RM), aims to validate and implement a non-intrusive vehicular traffic monitoring system based on computer vision and the Internet of Things (IoT), oriented towards its application in air quality studies. The methodology considered, in a first stage, the technical and economic evaluation of different capture technologies, selecting a Milesight industrial camera over other commercial alternatives and local prototypes due to its robustness, availability, and Edge Computing capabilities. Subsequently, object detection algorithms using convolutional neural networks (YOLO) and tracking techniques were implemented and calibrated. Their performance was validated through experimental campaigns comparing automatic counts with controlled manual records. This process allowed identifying and mitigating critical challenges associated with vehicular occlusion and the capture system's angle of incidence(...).
Description
Keywords
Flujo vehicular, Monitoreo de tránsito, Reconocimiento de imagenes, Dashboard, Visión por computador, YOLO (You Only Look Once)
