UCD-Where2comm: ground-prior-guided communication-efficient collaborative 3D object detection for multi-UAV swarms

Abstract

Multi-UAV collaborative perception enlarges the observation field and reduces occlusion by sharing complementary observations, but deployment is constrained by wireless bandwidth and by the difficulty of mapping camera features from different UAV heights, viewing angles, and baselines into a consistent 3D space. This paper proposes UCD-Where2comm, a camera-only multi-UAV collaborative 3D object detection framework that jointly addresses where to communicate and how to construct reliable bird's-eye-view (BEV) features before communication. The framework integrates three components: an adaptive confidence-to-binary communication module that transmits only budgeted and perceptually useful BEV cells, a Ground-Prior-Guided Feature Mapping (GFM) module that narrows the depth search range using calibrated ground priors, and a training-only Homologous-Point Geometric-Consistency Loss (HPL) that improves cross-view 3D agreement without inference overhead. Received sparse messages are warped into the ego frame and fused by confidence-aware attention. Experiments on CoPerception-UAVs show improved detection accuracy over single-UAV and Where2comm baselines with communication cost explicitly bounded by the spatial mask budget.