Robotic harvesting of Gerbera Jamesonii based on detection and three-dimensional modeling of cut flower pedicels

  • Authors:
  • Thomas Rath;Marco Kawollek

  • Affiliations:
  • Leibniz Universität Hannover, Institute of Biological Production Systems, Biosystems and Horticultural Engineering Section, Herrenhäuser Straíe 2, D-30419 Hannover, Germany;Leibniz Universität Hannover, Institute of Biological Production Systems, Biosystems and Horticultural Engineering Section, Herrenhäuser Straíe 2, D-30419 Hannover, Germany

  • Venue:
  • Computers and Electronics in Agriculture
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Within the present study, a system for the automated harvest of Gerbera jamesonii pedicels with the help of image analytic methods was developed. The study can be divided mainly into two parts: the development of algorithms for the identification of pedicels in digital images and the development of procedures for harvesting these pedicels with a robot. Images of plants were taken with a stereo camera system, which consisted of two high-resolution CCD-cameras with near-infrared filters. The plant was positioned on a rotatable working desk and images of eight different positions were shot. The developed image processing algorithm segmented the potential pedicel regions in the images, removed noise, differentiated overlapping pedicels by using different algorithms and combined the remaining regions to pedicel objects. From the data of both images and eight plant positions three-dimensional models of the pedicels were created by triangulation. The remaining parts of the plants were modeled in a simple fashion. The evaluated 3D model is used to calculate spatial coordinates for the applied robot control. For harvesting the pedicels, an industrial robot with six axes (plus an additional linear axis) was used. A pneumatic harvest grabber was developed, which harvested the pedicels by cutting them off. In order to guarantee the collision free path of the robot, a path planning module was integrated, which includes the three-dimensional model of the plant and the test facility. With the applied techniques it was possible to correctly detect all pedicels on about 72% of the images. Regarding the whole image series of the respective plant, all pedicels could be detected in at least one photographing position in 97% of all cases. In the harvest experiments 80% of all pedicels could be harvested. The harvest rates decreased with increasing numbers of pedicels on a plant. Therefore, 98% of the pedicels could be harvested of plants with one or two pedicels, but only 51% were harvested of plants with five or more pedicels. In horticultural practice, an identification system for evaluating the stage of maturity should be included. An implementation for harvesting pedicels of different species with similar basic characteristics is imaginable.