University of Western Ontario - Electronic Thesis and Dissertation Repository

Location of Thesis Examination

Room 320 Middlesex College

Degree

Master of Science

Program

Computer Science

Supervisor

Dr. John L. Barron

Abstract

As plant science advances, the quantitative measurement of 3D plant growth has become an essential tool to evaluate the growth performance of genetically modified seeds or plants. In the past, “wet volumes” had been used to quantitatively measure plant growth which result in the destruction of the plants. Other traditional methods like interferometry methods or mechanical methods are also inefficient because they are either invasive or unable to provide sufficient growth information of the plant body. On the other hand, computer vision methods, especially optical and range flow techniques, are non-contact, more accurate, and relatively less expensive. But they require a fixed light source for the experimental environment and also, higher acquisition rate for the data.

In this thesis, unlike optical or range flow methods, we generate a number of 3D range scans of the Arabidopsis thaliana plant from different viewpoints to reconstruct 3D polygonal meshes of it. We use ShapeGrabber range sensor to acquire the scanned data and Geomagic Studio 12 CAD software to register these scans to produce 3D polygonal meshes. Then, the canopy surface area and 3D stem volume of the plant are computed from these meshes to determine its growth over a time cycle. To perform the registration process in Geomagic, 6 ping pong balls are used as reference spheres.However, because the laser scanner can only see the parts of an object that are visible from its viewpoint, the balls are imaged as semi-spheres in the original range images.But, Geomagic is unable to register multiple images with incomplete semi-sphere data and requires full spheres as target objects. Zhao (MSc UWO 2010) and Yang (MSc UWO 2009) manually replaced these semi-spheres with artificially generated spheres using Geomagic which is both tedious and error prone. Moreover, they manually rotated the images to find out the pair of spheres that matches in adjacent range scans and aligned them manually before performing the registration. Because of the manual nature of this processing, automation of the registration process is very desirable. Among the major contributions of this research are not only to automatically generate synthetic sphere data but also detect and localize the semi-spheres in the original range images using parameter estimation techniques and parametric equation of spheres. Additionally, we reconstruct the original range images by replacing the incomplete semi-sphere data with “perfect” full sphere data to aid Geomagic during the automatic detection of target objects. After this pre-processing, the modified images are automatically registered using Geomagic macros so that the 3D polygonal meshes found from the registration process can be used to measure plant growth quantitatively. We believe that the automation of the registration process is a good first step towards a fully automated system of 3D plant growth measurement.