An Adaptive Feature Fusion Based Network for Aerial-Ground Point Cloud Place Recognition

Point cloud place recognition(PCPR) plays a key role in helping aerial-ground robotic systems navigate and understand their surroundings, especially when GPS signals are unavailable. However, the heterogeneity between point clouds collected by aerial and ground platforms poses challenges and severely affects the robustness and generalization of aerial-ground PCPR. New approaches are required to tackle the differences in how data from various sources are represented and to improve performance in challenging environments. To solve this issue, we introduce a hierarchical adaptive feature fusion network built on the feature pyramid network (FPN), which dynamically adjusts the importance of features at different scales, boosting those critical for place recognition tasks and minimizing the domain gap between diverse point clouds. The experimental results on benchmark datasets highlight the effectiveness of the proposed method.