A Visual Odometry System for Autonomous Vehicles Based on Squared Planar Markers Map

In the context of autonomous vehicles operating in small-scale environments such as industrial parks, residential areas, and university campuses, we propose a visual odometry system with a stereo camera based on a map of squared planar markers (SPM). Our contributions are primarily reflected in two aspects. Firstly, we have proposed a method for creating SPM maps. Compared to high-definition (HD) maps, our approach utilizes squared planar markers that are easier to create than road markings, traffic lights, and buildings, resulting in significant cost savings. Secondly, building on the foundation of stereo ORB-SLAM2, our proposed system accurately estimates the camera's position within the pre-existing SPM map by leveraging the pose information obtained from the planar markers. It also provides real-time outputs of latitude and longitude information. Furthermore, our system incorporates loop closure detection and correction capabilities, enabling the correction of camera pose by utilizing the relative poses between multiple planar markers. Extensive experimental evaluations demonstrate the promising performance of our system in outdoor scenarios, showcasing localization accuracy improvements of up to 30% compared to stereo ORB-SLAM2.