Online Terrain Mapping for Exploring Dense Forests on Unmanned Aerial Vehicles

This paper presents a novel method for online terrain mapping for unmanned aerial vehicles (UAVs) operating in dense forest environments. The proposed approach addresses the challenges of robust terrain height estimations for autonomous exploration using UAVs. By leveraging LiDAR data for ground segmentation, our method incrementally estimates terrain heights through a combination of RANSAC-based local plane fitting and a continuity test among local terrain planes. This approach ensures robust terrain mapping while UAVs navigate in dense forests, even in the presence of sparse or noisy LiDAR data. Experimental results in simulated and real forest environments demonstrate the effectiveness of our approach, showing significant improvements in terrain height estimation accuracy compared to baseline methods. The proposed system enables UAVs to maintain low-altitude flight relative to the terrain, facilitating successful exploration tasks like searching for a missing person in complex forests.