Self-localization of a mobile robot using monocular vision of a chessboard pattern

Abstract

A computationally fast visual approach of autonomous robot localization with high degree of accuracy is demonstrated. Relative position and orientation of an autonomous robot with respect to a fixed reference point is determined by minimal processing of a monocular image employing projective geometry resulting in very low computational requirements. A three-layered sequential algorithm i.e. Detection-Pixel Level Analysis-Sub Pixel Accuracy, is used to speed up the process and to reduce computational bulk. The position coordinate of a point in a plane is obtained by analyzing the perspective projection of a reference structure on a monocular image perceived by a camera placed at that point. A chessboard pattern is suitably chosen as the reference structure which is placed at the origin of the plane.