2D Positioning of Ground Vehicles using Stereo Vision and a Single Ranging Link

Abstract

In this work we propose a positioning method for ground vehicles in planar motion, based on sensor fusion of stereo cameras and sparse ranging measurements obtained from a wireless network. The proposed method is an alternative localization solution when Global Navigation Satellite System (GNSS) is unavailable, with notably low requirements on infrastructures. It does not require a database of landmarks and it works in single-link scenarios, i.e., at most one station reachable at any time. In theory, to estimate two dimensional position without ambiguity, at least three ranging anchors are required. However, in GNSS-denied environments, it is often difficult to achieve simultaneous connectivity to three wireless stations. We propose to apply visual odometry technique to estimate relative motion of the vehicle using stereo cameras, and fuse the vision system with a single ranging link. The sensor fusion method can resolve absolute position unambiguously if the vehicle sequentially connects to two stations with known coordinates. Furthermore, the accuracy of the estimated trajectory is improved by fusing both ranging and visual measurements.