Examinando por Autor "Vegas López, Andriz"

Mostrando 1 - 1 de 1
  • No hay miniatura disponible
    PublicaciónAcceso abierto
    Development of a software tool to implement a vision-based navigation method in low-textured environments
    (Universidad del Valle, 2025) Vegas López, Andriz; Bacca Cortes, Eval Bladimir; GRUPO DE INVESTIGACIÓN
    A software tool was developed to implement a vision-based navigation method in low-texture outdoor environments where GNSS systems may be unreliable. The system uses the natural horizon profile (skyline) as the primary visual feature for localization. The architecture integrates three modules: Configuration, Execution, and Report Generation. The Execution Module implements skyline detection through semantic segmentation, point cloud alignment via the Iterative Closest Point (ICP) algorithm, and fusion with GNSS data through an Extended Kalman Filter (EKF). As a comparison method and using monocular depth estimation with MiDaS, RTAB-Map was used to obtain an optimized trajectory as a pose graph. Field tests were conducted at Universidad del Valle. In FT0 (25 m trajectory, approximately 2,800 poses), a position RMSE of 0.369 m and an orientation RMSE of 12.41◦ were obtained. In FT1 (40 m trajectory, approximately 7,700 poses and more challenging illumination conditions), a position RMSE of 1.259 m and an orientation RMSE of 30.47◦ were obtained. The software was developed following Scrum methodology with a modular ROS-based architecture, a graphical interface for interactive experimentation and visualization. The results validate the feasibility of skyline-based localization as a complement to traditional navigation systems in specific field scenarios.