Self-Localization of a Mobile Robot using Statistics Values of Image Intensities

Abstract

This paper proposes an image based self-localization method of a mobile robot. Images are compressed for each column, and average and standard deviation of pixels in each column are used. Environmental data and observation data which are the compressed image data at registration and observation stage respectively are matched and the position of the robot is obtained. A simple and robust matching method based on a voting process is introduced. Search range for the matching is defined based on the Kalman filtering framework. The compression method is applied to omnidirectional images, and several experiments of self-localization of a mobile robot with omnidirectional images evaluate the proposed methods.