SLAM 3D basado en una cámara RGB-D y PCL
Date
2022-10-01
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
En el presente trabajo se desarrolla un algoritmo capaz de resolver el problema de localización y mapeo simultáneos (SLAM), obteniendo la odometría visual a partir de imágenes en colores con mapa de profundidad (RGB-D) de una cámara Microsoft Kinect bajo el entorno de simulación Gazebo. Para ello, se
propone un pipeline de registro que tiene como objetivo encontrar la mejor estimación de movimiento de
cuerpo rígido para mapear una imagen de profundidad en otra, asumiendo una escena estática tomada
por una cámara en movimiento. El pipeline propuesto se basa en nubes de puntos organizadas, esto es, que dichas nubes se presenten como matrices 2D. Aprovechando esto, se emplea una técnica para reducir las muestras extraídas, llamada normal space sampling, aumentando la probabilidad de que el registro converja al mínimo global. Los resultados obtenidos se asemejan a la trayectoria real simulada por el robot.
In the present work an algorithm capable of solving the problem of simultaneous location and mapping (SLAM) is developed, obtaining visual odometry from color images with depth map (RGB-D) of a Micro- soft Kinect camera under the environment of Simulation gazebo. For this, a registration pipeline is proposed that aims to find the best estimate of rigid body movement to map one depth image to another, assuming a static scene taken by a moving camera. The proposed pipeline is based on organized point clouds, that is, the- se clouds are presented in 2D matrices. Taking advantage of this, a technique is used to reduce the samples drawn, called normal space sampling, increasing the probability that the record will converge to the global minimum. The results obtained are similar to the real trajectory simulated by the robot.
In the present work an algorithm capable of solving the problem of simultaneous location and mapping (SLAM) is developed, obtaining visual odometry from color images with depth map (RGB-D) of a Micro- soft Kinect camera under the environment of Simulation gazebo. For this, a registration pipeline is proposed that aims to find the best estimate of rigid body movement to map one depth image to another, assuming a static scene taken by a moving camera. The proposed pipeline is based on organized point clouds, that is, the- se clouds are presented in 2D matrices. Taking advantage of this, a technique is used to reduce the samples drawn, called normal space sampling, increasing the probability that the record will converge to the global minimum. The results obtained are similar to the real trajectory simulated by the robot.
Description
Keywords
rgb, d, pcl, slam3d, slam, gazebo
Citation
Proyecciones Vol. 20 (2)
Collections
Endorsement
Review
Supplemented By
Referenced By
Creative Commons license
Except where otherwised noted, this item's license is described as openAccess