Search

Thin leaves, fine stems, self-occlusion, non-rigid and slowly changing structures make plants difficult for three-dimensional (3D) scanning and reconstruction - two critical steps in automated visual phenotyping. Many current solutions such as laser scanning, structured light, and multiview stereo can struggle to acquire usable 3D models because of limitations in scanning resolution and calibration accuracy. In response, we have developed a fast, low-cost, 3D scanning platform to image plants on a rotating stage with two tilting DSLR cameras centred on the plant. This uses new methods of camera calibration and background removal to achieve high-accuracy 3D reconstruction. We assessed the system's accuracy using a 3D visual hull reconstruction algorithm applied on 2 plastic models of dicotyledonous plants, 2 sorghum plants and 2 wheat plants across different sets of tilt angles. Scan times ranged from 3 minutes (to capture 72 images using 2 tilt angles), to 30 minutes (to capture 360 images using 10 tilt angles). The leaf lengths, widths, areas and perimeters of the plastic models were measured manually and compared to measurements from the scanning system: results were within 3-4\% of each other. The 3D reconstructions obtained with the scanning system show excellent geometric agreement with all six plant specimens, even plants with thin leaves and fine stems.

The last few years have seen an increasing interest in the development of high throughput phenotyping platforms (HTP) that allow the automated measurement of plant growth and structure. These platforms have utilised various imaging technologies, including fluorescence imaging, thermal imaging, and visible imaging. Since plants are structurally complex and inherently three-dimensional (3D), 3D imaging and reconstruction approaches have distinct advantages over 2D imaging when it comes to quantifying structural information (such as leaf angle distribution, leaf area etc.). High throughput phenotyping platforms, when combined with 3D image analysis, enable researchers to investigate complex functional traits related to plant structure, including responses to external and internal signals or perturbations. Moreover, image-based reconstruction techniques are able to produce complex structural models of both large and small individual plants, which retain colour information and are topologically coherent. In this paper, we present a computational workflow for analysing the growth of Arabidopsis thaliana rosettes over time using stereo reconstruction. Stereo reconstruction for plant phenotyping generally includes data collection and point cloud generation. In our approach, we not only use stereo image pairs to generate point clouds of the plants but also reconstruct mesh surfaces for further analysis. By applying a semi-global matching algorithm, 2.5D point clouds of Arabidopsis growing in trays with multiple plants were generated. Point clouds were then filtered and segmented in order to isolate single plants using clustering methods. A variety of surface reconstruction methods were then used to generate meshes from the point clouds. These meshes were analysed to quantify the plant dimensions and surface area in three dimensions. The segmentation was stable over time and was used to generate a time series for the growth of the individual plants over time. The advantages o