Three-Dimensional Reconstruction of a Dormant Tree Using RGB-D Cameras

Abstract

<abstract> <bold>Abstract.</bold> A faithful three-dimensional (3D) reconstruction of a dormant fruit tree is essential in designing a robotic pruning system. This research explored the feasibility of using RGB-D camera-based 3D sensing technology in reconstructing a 3D dormant tree in virtual space. For the feasibility study, a 3D visual sensing system that consisted of two RGB-D cameras was developed in the laboratory to capture images of a dormant tree from two orthogonal views. Indoor experiments were conducted in a laboratory setup, and the results showed that the system could reconstruct a 3D dormant tree virtually with sufficient accuracy for robotic pruning applications at a distance of about 1 m from the target tree. It also was found that the accuracy of the 3D reconstruction decreased as the cameras were moved farther from the target tree but that acceptable accuracy was still possible when the cameras were 3 m away from the target. The 3D information acquired about the tree structure from two viewpoints effectively reduced occlusions among the branches and allowed the generation of a more complete and accurate 3D reconstruction of the tree than was possible with a single view. Outdoor validation tests in a cherry orchard showed that the capability of the system to acquire tree images was affected adversely by direct sunlight but that it performed well in the evenings when there was less sunlight. Although the RGB-D cameras must be improved for effective use in the direct sunlight that is encountered in outdoor applications, this research demonstrated a great potential for using such cameras as 3D sensing components to support robotic pruning in orchards.