Multi-Sensor Fusion Localization of Indoor Mobile Robot

Abstract

Localization is critical for map building in visual SLAM (Simultaneous Localization and Map). Currently, accurate localization systems, such as Motion Capture, are expensive and, as many of them, not easy for re-configuration, a property essential for field robot test. This paper proposes a camera-odometry fusion method, which bases on a camera-marker system of low-cost and easy for re-configuration. The technology is based on odometers by combining two different sensor modules and PES using EKF (Extended Kalman Filter). A critical problem of EKF is the unknown PES (Position Estiamte System) variance, which is always set as a constant in previous works. In this paper, we solve this problem by using PES marker-pair, instead of a solo marker, to directly estimate the variance of PES localization. Experimental results in indoor environment demonstrate that the proposed approach substantially improves the localization accuracy of SLAM compared with PES only and odometry only. The position error is found to be less than 40mm of our system.