Authors:
D. Valiente
;
A. Gil
;
L. Fernández
and
O. Reinoso
Affiliation:
Miguel Hernández University, Spain
Keyword(s):
Visual SLAM, Omnidirectional Images.
Related
Ontology
Subjects/Areas/Topics:
Informatics in Control, Automation and Robotics
;
Mobile Robots and Autonomous Systems
;
Robotics and Automation
;
Vision, Recognition and Reconstruction
Abstract:
In this paper we focus on the problem of Simultaneous Localization and Mapping (SLAM) using visual information obtained from the environment. In particular, we propose the use of a single omnidirectional camera to carry out this task. Many approaches to visual SLAM concentrate on the estimation of the position of a set of 3D points, commonly denoted as visual landmarks which are extracted from images acquired at the environment. Thus the complexity of the map computation grows as the number of visual landmarks in the map increases. In this paper we propose a different representation of the environment that presents a series of advantages compared to the before mentioned approaches, such as a simplified computation of the map and a more compact representation of the environment. Concretely, the map is represented by a set of views captured from particular places in the environment. Each view is composed by its position and orientation in the map and a set of 2D interest points represe
nted in the image reference frame. Thus, in each view the relative orientation of a set of visual landmarks is stored. During the map building stage, the robot captures an image and finds corresponding points between the current view and the views stored in the map. Assuming that a set of corresponding points is found, the transformation between both views can be computed, thus allowing us to build the map and estimate the pose of the robot. In the suggested framework, the problem of finding correspondences between views is troublesome. Consequently, with the aim of performing a more reliable approach, we propose a new method to find correspondences between two omnidirectional images when the relative error between them is modeled by a gaussian distribution which correlates the current error on the map. In order to validate the ideas presented here, we have carried out a series of experiment in a real environment using real data. Experiment results are presented to demonstrate the validity of the proposed solution.
(More)