Self-Supervised Localisation between Range Sensors and Overhead Imagery

TL;DR

This paper introduces a self-supervised learning approach for localizing vehicles by matching ground range sensor data with overhead satellite imagery, overcoming modality differences without requiring precise ground truth.

Contribution

It presents a novel, cost-effective, self-supervised metric localization method that effectively bridges the gap between satellite images and ground sensors, including radar, across diverse datasets.

Findings

Method is robust across multiple datasets

Effective with millimetre wave radar data

Does not require metrically accurate ground truth

Abstract

Publicly available satellite imagery can be an ubiquitous, cheap, and powerful tool for vehicle localisation when a prior sensor map is unavailable. However, satellite images are not directly comparable to data from ground range sensors because of their starkly different modalities. We present a learned metric localisation method that not only handles the modality difference, but is cheap to train, learning in a self-supervised fashion without metrically accurate ground truth. By evaluating across multiple real-world datasets, we demonstrate the robustness and versatility of our method for various sensor configurations. We pay particular attention to the use of millimetre wave radar, which, owing to its complex interaction with the scene and its immunity to weather and lighting, makes for a compelling and valuable use case.