A Model-Based Approach to Recovering the Structure of a Plant from Images

Abstract

Abstract. We present a method for recovering the structure of a plant directly from a small set of widely-spaced images. Structure recovery is more complex than shape estimation, but the resulting structure estimate is more closely related to phenotype than is a 3D geometric model. The method we propose is applicable to a wide variety of plants, but is demonstrated on wheat. Wheat is made up of thin elements with few identifiable features, making it difficult to analyse using standard feature matching techniques. Our method instead analyses the structure of plants using only their silhouettes. We employ a generate-and-test method, using a database of manually modelled leaves and a model for their composition to synthesise plausible plant structures which are evaluated against the images. The method is capable of efficiently recovering accurate estimates of plant structure in a wide variety of imaging scenarios, with no manual intervention.

Evaluating physiological responses of plants to salinity stress

“…These systems are powerful because they provide high time and spatial resolution. Because of this high resolution, it is possible to develop more detailed models of growth (Ward et al , 2015) and to estimate relative growth rates (RGRs) (Berger et al , 2012), as has recently been achieved for rice (Hairmansis et al , 2014; Campbell et al , 2015). Importantly, these measurements also allow assessment of the shoot’s ion-independent component of salt toxicity, which involves the inhibition of shoot growth from the moment of salt imposition (Berger et al , 2012), before salt has had time to accumulate in the shoot and significantly affect the shoot’s function [Supporting Methodologies Section 2, eqn (b)].…”

“…Few studies have reported root imaging under natural conditions and without destructive harvesting, such as imaging of roots using the Growth and Luminescence Observatory for Roots (GLO-Roots) system (Rellán-Álvarez et al , 2015) or in a more artificial system using transparent growth media (such as gel or glass beads) (Courtois et al , 2013; Topp et al , 2013). Improvements to non-destructive analyses involve not only automated data gathering and data analyses, but also the development of new technologies that allow determination of root parameters as well as the development of models to recover the structures of plants using 3D models (Ward et al , 2015). We believe that these technologies will provide a step change in salinity research, especially when time resolution is incorporated to provide insights into the dynamic responses of plants to salinity.…”

“…Consequently, to estimate tissue tolerance, measurements of senescence need to be made specifically in older leaves. One way to do this is to use image analysis models to separate individual leaves and to therefore enable measurements of parameters of each single leaf, rather than just those of the whole shoot (Ward et al , 2015). In effect, a full life history of each leaf could be developed, and the effects of salinity and genetic composition on that life history could be quantified.…”

A Robotic Platform for Corn Seedling Morphological Traits Characterization

“…However, it has proven to be challenging for stereo vision to handle the complexity of plant canopies given the difficulties in stereo matching caused by leaf occlusion and the lack of leaf surface texture. A visual hull algorithm was used to reconstruct 3D models of corn and barley plants [7], but it was found challenging to apply it to complex plant canopies [8]. Recently, plant phenotyping has gained more attention because of the development of advanced sensors and robotic data collection and plant monitoring methodologies.…”

“…In their approach, 3D reconstruction of plants was the first step to provide morphological and position information, which was needed to accomplish other operations on plants such as imaging, probing and cutting at specific locations. Alenyà et al [12] used ToF depth data to perform quadratic surface fitting that was applied to segment plant images [8]. They showed that the obtained surface fit well with the target leaves and the candidate leaves could be approached by a robot-mounted camera using location information.…”