Deriving A Priori Environment Information for Ground-Based Exploration Using Aerial Observations

Autonomous Mobile Robots (AMRs) are emerging as a potential solution for transitioning agriculture towards sustainability by optimizing resource utilization in applications such as water and fertilizer use. However, deploying AMRs requires a map of the environment for effective operation.
This study investigates UAVs equipped with RGB cameras to derive such a map that can be used by ground-level AMRs in vineyards. Our approach includes capturing detailed and frequent aerial imagery to reflect the vineyard's condition, identifying obstacles and separating vine growth from other vegetation through terrain model-based segmentation, and refining segmentation through vine-row reconstruction to enhance the accuracy and utility of the map for navigation. We compare our approach on three flights on a local vineyard with manually annotated ground truth data and show that our approach is a viable solution to derive an a priori map for autonomous ground vehicles in vineyards.