Three-dimensional reconstruction of plant shoots from multiple images using an active vision system

Abstract

The reconstruction of 3D models of plant shoots is a challenging problem central to the emerging discipline of plant phenomics – the quantitative measurement of plant structure and function. Current approaches are, however, often limited by the use of static cameras. We propose an automated active phenotyping cell to reconstruct plant shoots from multiple images using a turntable capable of rotating 360 degrees and camera mounted robot arm. To overcome the problem of static camera positions we develop an algorithm capable of analysing the environment and determining viewpoints from which to capture initial images suitable for use by a structure from motion technique.