Robust Sensor Fusion with Pairwise Dynamic Covariance Scaling for Localization in Urban Areas

We propose a robust sensor fusion method that is for vehicle localization in urban areas. Our main contribution is to merge the absolute position measurements from the vision-based localization module and the consumer level GPS via suggesting a robust back-end, named Pairwise Dynamic Covariance Scaling (PDCS). PDCS is applicable even when the measurement uncertainty is unknown, which is possible because we let the system considers different sensor measurements as a pair. Specifically, we penalize measurements with large relative errors compared to their paired counterparts, instead of penalizing measurements with large absolute errors. As seen from the experiment in the real urban area, the proposed method is less sensitive to user parameters as verified by its steady localization performance over the varying parameters.