Propuesta metodológica para la identificación de señales de tránsito horizontales en ciclorrutas, caso práctico parque El Tunal, Bogotá.

Abstract

This monograph develops an artificial intelligence-based methodological proposal for the automated identification of horizontal traffic signs on urban cycle paths, using El Tunal Park in Bogotá as a case study. The proposed methodology includes data capture in geolocated video format, frame preprocessing, training of a detection model using the YOLOv5n architecture (nano version), and its subsequent evaluation using geospatial tools. The workflow integrates technical components of computer vision and spatial analysis, making use of the Full Motion Video feature of ArcGIS Pro to validate the model's performance in the terrain. The results show that the model correctly identifies several horizontal signs, although it also reveals areas with missing or deteriorated signage, demonstrating the tool's usefulness for urban diagnostic processes. Opportunities for improvement are identified in the quality and quantity of the dataset used for training, as well as in the diversification of the evaluated environments, in order to strengthen the model's generalization capacity. The main conclusion is that the proposed methodology is viable, replicable, and provides an innovative approach to road sign monitoring, combining artificial intelligence and geospatial analysis applied to bicycle infrastructure management.