Road Similarity-Based BEV-Satellite Image Matching for UGV Localization

TL;DR

This work tackles the problem of precise UGV localization in GNSS-denied off-road environments by proposing a matching-based framework that aligns a locally generated BEV perception image with a satellite map in a shared road similarity space. A high-precision LiDAR-Inertial odometry backbone feeds multi-modal data to create a BEV, which, together with a satellite map, is processed in a road similarity space using a U‑Net-derived feature representation and online clustering. Localization is performed via NCC-based image matching within a particle filter, complemented by global path alignment to refine pose, yielding robust long-range localization even under nighttime conditions. The approach demonstrates superior accuracy and stability over a 5.2 km trajectory and a 10 km test, with notable robustness to seasonal and lighting variations, offering practical benefits for GNSS-denied autonomous navigation.

Abstract

To address the challenge of autonomous UGV localization in GNSS-denied off-road environments,this study proposes a matching-based localization method that leverages BEV perception image and satellite map within a road similarity space to achieve high-precision positioning.We first implement a robust LiDAR-inertial odometry system, followed by the fusion of LiDAR and image data to generate a local BEV perception image of the UGV. This approach mitigates the significant viewpoint discrepancy between ground-view images and satellite map. The BEV image and satellite map are then projected into the road similarity space, where normalized cross correlation (NCC) is computed to assess the matching score.Finally, a particle filter is employed to estimate the probability distribution of the vehicle's pose.By comparing with GNSS ground truth, our localization system demonstrated stability without divergence over a long-distance test of 10 km, achieving an average lateral error of only 0.89 meters and an average planar Euclidean error of 3.41 meters. Furthermore, it maintained accurate and stable global localization even under nighttime conditions, further validating its robustness and adaptability.

Figures (9)

• Figure 1: A diagram of our localization system. The localization is achieved by projecting the BEV perception image, generated through multi-sensor spatiotemporal fusion, and the satellite map into the road similarity space for matching.
• Figure 2: The bird’s-eye view generation process: For each frame of temporally synchronized point clouds and RGB images, colorize the point clouds using a projection matrix, and accumulate multiple frames of point clouds relying on odometry to complete the bird’s-eye view generation.
• Figure 3: The road similarity analysis process: First, the feature map is obtained through model inference, and the features of traversable regions are extracted using recorded odometry data. Then, the prototype vectors are analyzed through online clustering, and finally, the traversability map is generated by computing with the feature map.
• Figure 4: Global satellite road similarity map.
• Figure 5: Image Matching Process: The BEV perception image and the satellite map undergo road similarity analysis. Each particle extracts a satellite map patch from the global road similarity map, using the prior position estimated from odometry as a reference. The NCC between the satellite patch and the BEV road similarity image is then computed to determine the matching score for each particle, where redder colors indicate higher matching scores.