Abstract
High-Definition (HD) maps play a crucial role in autonomous driving systems.Recent methods have attempted to construct HD maps in real-time based oninformation obtained from vehicle onboard sensors. However, the performance ofthese methods is significantly susceptible to the environment surrounding thevehicle due to the inherent limitation of onboard sensors, such as weakcapacity for long-range detection. In this study, we demonstrate thatsupplementing onboard sensors with satellite maps can enhance the performanceof HD map construction methods, leveraging the broad coverage capability ofsatellite maps. For the purpose of further research, we release the satellitemap tiles as a complementary dataset of nuScenes dataset. Meanwhile, we proposea hierarchical fusion module that enables better fusion of satellite mapsinformation with existing methods. Specifically, we design an attention maskbased on segmentation and distance, applying the cross-attention mechanism tofuse onboard Bird's Eye View (BEV) features and satellite features infeature-level fusion. An alignment module is introduced before concatenation inBEV-level fusion to mitigate the impact of misalignment between the twofeatures. The experimental results on the augmented nuScenes dataset showcasethe seamless integration of our module into three existing HD map constructionmethods. It notably enhances their performance in both HD map semanticsegmentation and instance detection tasks.