Image-based tree pruning

Abstract

There is an increasing awareness and development of agricultural robots to take the toil of farming by automating growing plants and trees. Pruning is an expensive and labor intensive step in growing trees, that greatly affects its productivity. Moreover, pruning requires knowledge about what, where and how to cut. To partially solve the limitations of manual pruning methods, this paper presents an automatic image-based pruning system. Our system uses a high-resolution and a Kinect camera mounted on a mobile robot to capture the 3D structure of trees in the field. The robot goes around a tree and synchronously captures high-resolution and depth images. The visual and depth information across images is fused to estimate a 3D “stick” representation of the tree. The output of our system suggests the operator which branches to cut based on pre-existing rules. Several challenges contribute to the difficulty of image-based pruning: (1) fusing spatial and temporal information in depth images, (2) capture and segment small branches, (3) quantitative estimation of the angles and length for each branch. The number of suggested branches to cut in several trees have high agreement with the ones suggested by an expert, that illustrates the validity of our approach.