Collaborative UAV-UGV Environment Reconstruction in Precision Agriculture

In air-ground multi-robots applications, where both Unmanned Aerial Vehicles (UAVs) and Unmanned Ground Vehicles (UGVs) operate in a coordinate way, the ability to obtain a unified environment representation of the target area is an essential requirement. However, a global registration of heterogeneous ground and aerial maps is a challenging task, especially for agricultural scenarios: the visual appearance of such kind of environment is rather homogeneous, it is difficult to find and exploit distinctive 3D geometrical structures, while the maps built using robots of different types show differences in both size and resolution and, possibly, scale errors. In this paper, we tackle the cooperative UAV-UGV environment modeling problem in farming scenarios. We propose a novel maps registration pipeline that leverages a digital multi-modal environment representation which includes a vegetation index map and a Digital Surface Model (DSM). Using such map representation, we cast the data association problem between maps built from UAVs and UGVs as a multi-modal, large displacement dense optical flow estimation. The data association is then used to estimate an initial, non-rigid alignment between the maps that also compensates the (directional) scale discrepancies between them. A final refinement is then performed, by exploiting only meaningful parts of the registered maps. We compare our method with standard registration techniques showing better alignment performances and better generalization properties over different misalignments and scale errors.