Análisis del flujo vehicular en imágenes aéreas por medio de técnicas de visión artificial

Abstract

Abstract: Vehicular traffic in the city of Loja is currently a daily challenge in urban areas, especially exacerbated during rush hour. In order to address this problem, a prototype was developed based on computer vision techniques for vehicle flow recognition. To carry out this research, a proprietary dataset was generated by capturing images from a drone in the aforementioned city. This dataset was used for the classification of different types of vehicles, such as cars, motorcycles, heavy vehicles and buses. The Roboflow tool was used for image labeling. Subsequently, the model was trained using the YOLOv8 neural network, achieving satisfactory results in terms of object recognition, with mAp (mean Average Precision) values of 80.3%. These data are presented in an accessible way for users through a web platform and a mobile application, facilitating the visualization and understanding of information related to vehicular flow in the city.