CETM: Cooperative Evidence Trust Management for Aerial-Ground Cooperative Perception

Aerial-ground cooperative 3D perception fuses complementary observations from ground vehicles and unmanned aerial vehicles, offering a promising paradigm for enhancing automated-driving perception, extending sensing coverage, and alleviating occlusion. However, aerial-ground deployment is constrained by two key bottlenecks, namely cross-platform observation heterogeneity that leads to inconsistent spatial evidence among candidates and communication temporal perturbations that tend to cause unstable tracking identities. To overcome these limitations, we propose Cooperative Evidence Trust Management (CETM), an online receiver-side management framework for aerial-ground cooperative perception. Specifically, the framework contains two complementary modules, where Spatial Reliability Calibration (SRC) calibrates candidate ranking using cross-route spatial evidence, while Lineage-Guided Memory (LGM) maintains stable scene-local identities using spatiotemporal lineage evidence. Extensive experiments on the Griffin dataset demonstrate that CETM achieves state-of-the-art cooperative detection and tracking performance, outperforming existing methods by a large margin and exhibiting strong robustness to communication perturbations. The code and model will be publicly available.