Landmark Assisted Stereo Visual Odometry

Abstract

In this paper, we consider the positioning technology to facilitate the system services for automatic guided vehicle. Typically, quick response (QR) code is a feasible solution with high accuracy. However, the position information of QR code is discrete and sparse, which leads to uncertainty in position interpolation. To solve this problem, we proposed a positioning system of feature-based stereo visual odometry (VO) by combining some discrete and sparse absolute positioning information. In the proposed system, we adopt stereo cameras to obtain the depth of feature points directly and then apply an algorithm with two parallel threads of tracking and mapping. The mapping thread deals with not only information sent by tracking thread but also those sent by QR code. This proposed system has been tested and verified on our automatic guided vehicle. The measured and computational results suggest that the proposed system has acceptable robustness and relatively high accuracy.