Eventos Anais de eventos
COBEM 2019
25th International Congress of Mechanical Engineering
MAPPING IN DYNAMIC ENVIRONMENTS USING DEEP LEARNING-BASED HUMAN DETECTION
Submission Author:
João Carlos Virgolino Soares , RJ
Co-Authors:
João Carlos Virgolino Soares, Marcelo Gattass, Marco Antonio Meggiolaro
Presenter: João Carlos Virgolino Soares
doi://10.26678/ABCM.COBEM2019.COB2019-0845
Abstract
Simultaneous Localization and Mapping (SLAM) is one of the key problems in mobile robotics. However, most of the state-of-the-art SLAM systems only work properly in static environments, limiting their applicability. Performing SLAM in scenarios with dynamic objects is still an open problem, with solutions usually relying on Optical Flow or feature-based methods. Deep learning techniques offer a high-level information of the scene that can improve the solution in environments with a priori dynamic objects, such as people. This work proposes an extension of RTAB-Map, a state-of-the-art SLAM system, to enable mapping in the presence of people in the scene using a Deep Learning-based Object Detection algorithm. The system is evaluated with an RGB-D dataset and an experimental setup.
Keywords
SLAM, Dynamic Environments, Deep learning, Object Detection, ROS
