Towards an optimal pipeline for plant point cloud generation using a low-cost hardware and software approach

Plant phenotyping is a time-consuming task that can be automated with 3D scanning technology. The goal is to create 3D plant point clouds at low cost. Photogrammetry, specifically the structure-from-motion/multi-view stereo (SfM/MVS) method, is particularly suited for this purpose. The method generates 3D point clouds from 2D images by determining and refining camera positions and 3D structures from overlapping images. To completely capture a plant, a turntable and an automated camera system consisting of four 16MP autofocus cameras are developed. An Arduino Uno microcontroller controls the rotating mechanism. The influence of different backgrounds (black, white, gray, cyan, magenta, yellow) is investigated in terms of improving image quality and 3D reconstruction, respectively. For image processing three open-source software are compared and the resulting 3D point clouds are evaluated oriented to the VDI 2634/1 guideline.