Using Global Appearance Descriptors to Solve Topological Visual SLAM

Abstract

Nowadays, mobile robots have extended to many different environments, where they have to move autonomously to fulfill an assigned task. With this aim, it is necessary that the robot builds a model of the environment and estimates its position using this model. These two problems are often faced simultaneously. This process is known as SLAM (Simultaneous Localization and Mapping) and is very common since when a robot begins moving in a previously unknown environment it must start generating a model from the scratch while it estimates its position simultaneously. This work is focused on the use of computer vision to solve this problem. The main objective is to develop and test an algorithm to solve the SLAM problem using two sources of information: (a) the global appearance of omnidirectional images captured by a camera mounted on the mobile robot and (b) the robot internal odometry. A hybrid metric-topological approach is proposed to solve the SLAM problem.