Master Thesis

Development of a surface model for the characterization of plants leaves for grasping and next-best-view planning

Work default illustration

Supervisor/s

Information

  • If you are interested in the proposal, please contact with the supervisors.

Description

The EU project GARNICS (''Gardening with a cognitive system'') deals with the phenotyping of plants using image processing techniques. In this context, we need to develop appropriate 3D models of plants and plant substructures, i.e. leaves. The goal of this student project is the development of an a parametric description of surface structures of Tobacco plant leaves using sparse depth information acquired via a PMD camera. The model should allow for the selection of grasp points or planning of next best views for data completion.

1. Acquisition of plant (leaves) images (color + depth) with an RGB and a PMD camera.
2. Initially manual or half-manual segmentation of plant leaves using color images.
3A. Development of an parametric surface leaf surface model using the acquired set of images that delivers a “good” description of the data while using as few parameters as possible – this can be formulated as an optimization problem.
(3B.) Leaves having complex shapes may be better described by a relational graph of surface patches, where each surface patch represents a parametric surface models (a leaf may be modeled by two planar surface patches enclosing an angle ß).
4. Learning of model parameters using the data set.
5. The parametric surface model will be combined with image segmentation to automatically segment plant leaves from images using an existing framework.
6. Based on the surface models, grasps or new viewpoints are suggested and executed in robotic experiments.