3D Reconstruction, Modelling and Analysis of in Situ Root System Architecture

Root system architecture (RSA) plays an important role in water and nutrient uptake for plant development and growth and hence in grain yield. In situ studies of RSA root architecture will assist in the characterization of phenotypes for the purpose of identifying genotypes of cereal plants that are more stress tolerant and are producer of higher grain yield. In this paper we present a method of 3D reconstruction of roots with a non-destructive method, preserving the RSA and also allowing for observation of the roots growth as function of time. Our method was applied to corn plant seedlings in their early stage of growth. The germinated corn seeds are grown in a transparent gel-based growth system. Observations are made with a digital camera with the growth chamber sitting on a turntable, at different time instances of the growth. The digital images are processed for segmentation of roots, root tip detection, and root tips are tracked as a function of angle of the turntable motion. The elliptical trajectories of the root tips are used to calibrate the cameras. After calibration of the imaging sensor the 3D reconstruction and modelling of the roots are done using the visual hull (space carving) algorithm We analyse different stages of 3D root developments in terms of roots volume as a function of time and the number of images used in reconstruction of the RSA. Our results show that root biomass/volume increases with time. This trend is true irrespective of the number of images used for 3D reconstruction. The increase in root volume is approximately linear with time for plants is in the early stage of growth. It is not very clear what will be the optimal number of images to use in 3D reconstruction and modelling. Less number of images require less acquisition time and less processing time, which may be important for high throughput systems. By visual inspection it can be said that the 3D reconstruction becomes more accurate and detailed with increasing number of images. We empirically verify that the volume of reconstruction monotonically decreases increasing number of images from different viewpoints. This is due to erroneous voxels being carved out when information from more images are being added to 3D reconstruction.