Reconstructing the Depth of Hidden Branches - New IEEE RA-L Paper
For drones flying in forests branches present dangerous obstacles, in this work we show that we can reconstruct the depth of also occluded branches using a neural network.
When drones navigate through forests, detecting and avoiding vegetation becomes crucial. While leaves are soft and pose little harm to the drone, they conceal rigid branches that can pose a concrete danger and need to be avoided.
In a recent study published in IEEE RAL, we successfully trained a neural network to infer the structure of partially hidden tree branch from RGB-D images. This marks a significant first step towards enhancing safety during flights through dense vegetation, opening up new possibilities for environmental monitoring and precision agriculture.
Kudos to Christian Geckeler, Emanuele Aucone, Yannick Schnider, Andri Simeon, Jan von Bassewitz-Philipp von Bassewitz and Yunying Zhu for their excellent work. Special thanks to Zoo Zürich for granting us the opportunity to validate our neural networks in the Masoala Rainforest Halle.