Automated reconstruction of 3D plant architecture applied to grapevine phenotyping

Abstract

We propose an approach to automated 3D reconstruction of plant architecture followed by a precise derivation of phenotypic traits. We applied and evaluated our approach within a grapevine phenotyping project. The reconstruction results and the derived phenotypic traits could be shown to be in line with given ground truth data. Furthermore, our approach that combines a generative modelling of plants with a probabilistic sampling of the space of reconstruction hypotheses seems to be a promising framework that could potentially improve automated plant reconstruction and automated phenotyping in a broader field of applications.