A FastSLAM-based algorithm for omnidirectional cameras

Abstract

Environments with a low density of landmarks are difficult for vision-based Simultaneous Localization and Mapping (SLAM) algorithms. The use of omnidirectional cameras, which have a wide field of view, is specially interesting in these environments as several landmarks are usually detected in each image. A typical example of this kind of situation happens in indoor environments when the lights placed on the ceiling are the landmarks. The use of omnivision combined with this type of landmarks presents two challenges: the data association and the initialization of the landmarks with a bearing-only sensor. In this paper we present a SLAM algorithm based on the wellknown FastSLAM approach. The proposal includes a novel hierarchical data association method based on the Hungarian algorithm, and a delayed initialization of the landmarks. The approach has been tested on a real environment with a Pioneer 3-DX robot.