Detalle de la tesis

Autores: Pire, Taihú Aguará Nahuel.

Resumen: In order to allow a mobile robot navigate and perform tasks autonomously, it must know its pose (position and orientation) and have a representation of the environment (map). In environments where the robot does not have a previous map and no external information is provided to know its pose, it is necessary to perform both tasks simultaneously. The problem of localizing a robot and building a map of the environment simultaneously is called SLAM; this stands for Simultaneous Localization and Mapping.In this thesis, a system based on stereo vision to address the problem of SLAM is presented. The method, called S-PTAM as an acronym for Stereo Parallel Tracking and Mapping, was developed. This method is intended to run in real-time for long trajectories, allowing to estimate the pose accurately as it builds a sparse map of the environment on a global coordinate system.For optimal performance, S-PTAM decouples localization and mapping tasks of the SLAM problem into two independent threads, allowing us to take advantage of multicore processors. Besides the localization and the mapping modules, a loop closure module that can recognize places previously visited by the robot is proposed. The detected loops are used to refine the map and the estimated trajectory, effectively reducing the accumulated error of the method so far.S-PTAM works on the visual features extracted from the images provided by the stereo camera. To determine which feature extractor is the most suitable in terms of accuracy, a comparison in terms of robustness and computational cost of the most relevant detectors and binary descriptors in the literature is performed.Finally, experiments with public datasets for validating the accuracy and performance of the proposed method are presented. As a result S-PTAM is one of the most accurate SLAM methods of the state of the art. S-PTAM was released as free software to ease its use and to allow comparison with other SLAM methods.

Grado académico: Universitario de posgrado/doctorado.

Titulo obtenido: Doctor en Ciencias de la Computación.

Idioma: Inglés.

Area de conocimiento: Control Automático y Robótica.

Año: 2017

tesis_web: Enlace a tesis