Skip to Main content Skip to Navigation

Modélisation 3D et suivi visuel pour caractériser le phénotype de variétés de tournesol

William Gélard 1
1 LAAS-RAP - Équipe Robotique, Action et Perception
LAAS - Laboratoire d'analyse et d'architecture des systèmes
Abstract : The constant increasing food and energy demand in the world associated to global warming and climate change issues, pushed the researchs in plant breeding to move towards the improvement of crops performance and development of a more sustainable agriculture. To meet these demands, the effort made by the researchers were focused on the development of high-throughput genotyping methods (i.e., the study of genome sequence of plants) and allowed the biologists to indentified the genotypes of a large amount of plants. Moreover, understanding the relationships that link the genotypes (DNA) to the phenotypes (visual characteristics) that evolve according environmental conditions like : light, water, drought, heat, etc. has become a main issue in agricultural research. While the genotyping methods were rapidly improved and automatized during the last decade, the phenotyping methods remain manual, sometimes destructive and non-replicable. The usual phenotyping methods consist to measure certain visual parameters of a plant such as : main stem heigh, number of leaves, leaf initiation angle or leaf area, but more importantly, be able to follow these parameters along the plant growth. Consequently, the number of plants to harvest is very important and the measurements are extremely time-consuming. The emergence and reliability of new technologies in computer vision and robotic have led the researchers to take an interest in them and to seek how they can be used in plant science. The thesis is focused on the design, development and validation of a high-throughput phenotyping method design for sunflower plant with an eye to amplify phenotyping capacities by Agronomists and Geneticists (and later varieties evaluators and seed producers). The aim is to improve, modernize and automatize the current phenotyping methods as a way to help the plant scientists to collect a large amount of data. Motivated by the wish to perform high-throughput plant phenotyping, we propose a 3D approach to automatically extract visual characteristics of sunflower plants grown in pot. First, a 3D point cloud of a plant is acquired with classical SfM techniques. A segmentation step is then proceeded to retrieve the main stem and the leaves. With the intention of following the visual characteristics during the plant growth, especially, the leaf area expansion rate of each leaf, a labelling step relying on the botanical model of a plant is performed to affect them a unique label that will not change over time. Finally, the visual characteristics are extracted and results obtained on sunflower plants demonstrate the efficiency of our method and make it an encouraging step toward high-throughput plant phenotyping.
Document type :
Complete list of metadata

Cited literature [129 references]  Display  Hide  Download
Contributor : Laas Hal-Laas <>
Submitted on : Thursday, June 27, 2019 - 3:19:17 PM
Last modification on : Tuesday, June 22, 2021 - 3:49:50 AM


GELARD William.pdf
Files produced by the author(s)


  • HAL Id : tel-02167190, version 1


William Gélard. Modélisation 3D et suivi visuel pour caractériser le phénotype de variétés de tournesol. Automatique / Robotique. Université Toulouse 3 Paul Sabatier (UT3 Paul Sabatier), 2018. Français. ⟨tel-02167190⟩



Record views


Files downloads