Conteo automatizado de flujo vehicular en imágenes aéreas usando herramientas deep learning

Abstract

Abstract:In the city of Loja, traffic flow is a recurring problem, which is exacerbated during rush hour. This work addresses this issue through an automated vehicle counting system based on computer vision and deep learning. Using drones, images and videos were collected in various urban conditions, creating a customized dataset with more than 700 labeled images and processed using the Roboflow tool. The model was trained with YOLOv5 and YOLOv8 deep learning technologies, increasing the mAP (mean average precision) to over 80%, ensuring high accuracy in the detection and classification of vehicles, such as cars, motorcycles, buses, and heavy goods vehicles. This system allows visualization of vehicle flow, providing support for decision-making in traffic management. Furthermore, the main contribution is shown in the implementation of accessible and precise technologies to improve urban mobility as a projection for the development of future intelligent transportation applications.