BRANCH - a Labeled Dataset of RGB-D Images and 3D Models for Autonomous Tree Pruning

Abstract

In autonomous pruning, the robot must identify the branches that need to be removed to improve the health and shape of the tree. Furthermore, the robot needs depth information about these branches in order to successfully perform tool navigation and finally pruning. In this work, we provide the BRANCH dataset with 70 RGB-D images of pear trees before and after pruning, taken in the real environment from different viewpoints to cover the whole tree. Based on these images, we created point clouds and performed model reconstruction to obtain 3D models of the trees. After overlaying the models before and after pruning, we obtained the points of the pruned branches. Therefore, we also provide labeled point clouds of 52 trees in BRANCH that can be used for further machine learning.