Predicción de intenciones de conductores en rotondas no estructuradas mediante detección de vehículos con YOLO y análisis estadístico

This study explores the complexity of predicting driver intentions in unstructured intersections, such as roundabouts without traffic signs or lane markings. These scenarios pose unique challenges for advanced driver assistance systems (ADAS) and autonomous vehicles. Unlike highways with well-define...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Vázquez, Damian Raimundo, Torres, Carlos, Mariguetti, Jorge A., Gramajo, Sergio, Robledo Sánchez Alberto, Andrés
Formato: Artículo revista
Lenguaje:Español
Publicado: Facultad de Ciencias Exactas y Naturales y Agrimensura - Universidad Nacional del Nordeste 2024
Materias:
Acceso en línea:https://revistas.unne.edu.ar/index.php/eitt/article/view/8238
Aporte de:
Descripción
Sumario:This study explores the complexity of predicting driver intentions in unstructured intersections, such as roundabouts without traffic signs or lane markings. These scenarios pose unique challenges for advanced driver assistance systems (ADAS) and autonomous vehicles. Unlike highways with well-defined lanes and traffic signals, unregulated roundabouts require a more detailed analysis of vehicle behavior. The proposed approach uses the YOLO object detection model to identify vehicles in a roundabout, focusing on specific areas like entrances and exits, rather than analyzing the entire scene, which improves accuracy and efficiency. A centroid-based tracking system is also implemented to avoid counting the same vehicle multiple times. Six zones are defined within the roundabout: three to predict driver behavior and three to count vehicles exiting the roundabout. The system also measures congestion time when vehicles remain stationary for a certain period, providing key information for traffic management. Results show a significant accuracy rate in predicting vehicle trajectories, though there are cases where predictions do not fully match actual vehicle movements, highlighting the need for algorithm improvements. The study also suggests that integrating machine learning models in the future could significantly enhance system performance. Ultimately, the research presents an innovative approach to improving traffic safety and efficiency in roundabouts, despite certain limitations such as the video capture angle.