HD-Map-aided Pipeline for Absolute Visual-Based Positioning

For decades, visual-based positioning solutions for autonomous vehicles have attracted research interest in the autonomous vehicle driving domain. As high-definition maps (HD-Maps) are gaining importance, maturity, and availability, the need to effectively incorporate HD-Map data into the positioning models has been increasing. The addition of larger image datasets with more manually labeled samples and more labels for different object classes has resulted in an abundance of highly accurate geo-referenced image datasets to play a significant role in solving the visual-based positioning problem. In this paper, we propose an HD-Map-aided vision-based positioning method that will offer the autonomous driving industry a revolutionary alternative to the mainstream GNSS-based positioning. With an update rate of 2.66s on average and global coordinates positioning mean absolute error of 1.0m. Compared to the 1s/4.0m of the commercial GNSS solutions, we claim to introduce a GNSS-free robust global positioning within pre-mapped areas.