Creación de un modelo visual del entorno basado en imágenes omnidireccionales

Abstract

This paper deals with the problem of Simultaneous Localization and Mapping (SLAM). The solution presented is based on the utilisation of a set of images to represent the environment. In this way, the estimation of the map considers the computation of the position and orientation of a set of omnidirectional views captured from the environment. The proposed idea sets apart from the usual representation of a visual map, in which the environment is represented by a set of three dimensional points in a common reference system. Each of these points is commonly denoted as visual landmark. In the case presented here, the robot is equipped by a single omnidirectional visual sensor that allows to extract a number of interest points in the images, each one described by a visual descriptor. The map building process can be summed up in the following way: as the robot traverses the environment, it captures omnidirectional images and extracts a set of interest points from each one. Next, a set of correspondences is found between the current image and the rest of omnidirectional images existing in the map. When the number of correspondences found is enough, a transformation is computed, consisting of a rotation and a translation (up to an unknown scale factor). In the paper we show a method that allows to build a map while localizing the robot using these kind of observations. We present results obtained in a simulated environment that validate the proposed idea. In addition, we present experimental results using real data that prove the suitability of the solution.