Plants are the major organisms that
can absorb the light energy
from the sun to produce biomass and oxygen. One key problem in studying
growth is to understand the photosynthetic activities
of plants under various external stimuli or genetic variations.
Because leaves at different developmental ages may response to the
change of the environmental conditions and gene mutations in very
different ways, it is important to conduct a leaf-level analysis
of the photosynthetic efficiency. Motivated by the needs in plant
biology, given images and videos captured by visual sensors, the goal
of the plant vision project is
to develop advanced computer vision
algorithms to automatically, accurately and efficiently estimate the structure of a plant, which includes
2D multi-leaf alignment and tracking, and 3D reconstruction of all
leaves in a plant.
have developed a framework based on the well-known Chamfer Matching
The input to our system is a fluorescence image or video of a plant,
is captured in a growth chamber (Fig. 1). Multi-leaf alignment
aims to segment/align all leaves with pre-defined leaf templates and
estimate the two
tip points of each leaf. Multi-leaf tracking aims
track all leaves over time based on the alignment results of one
frame. The leaf alignment and tracking results can directly benefit the
leaf behavior in plant biology, such as leaf growth, leaf-level
photosynthesis, leaf-level variations in plant mutant, etc. We have
recently extended our system to process RGB videos of plants as well.
Fig. 1 Overview of leaf alignment and tracking system
As shown in Fig. 2, the multi-leaf tracking algorithm consists of two steps. Firstly, a set of templates are applied to the test image to generate the same amount of leaf candidates. Secondly, we develop a multi-objective optimization process to select a subset of leaf candidates. The objective is to select a minimal number of leaf candidates with smaller Chamfer distances to cover the test image mask as much as possible.
Fig. 2 Overview of multi-leaf alignment
Multi-leaf tracking is an extension of the leaf alignment algorithm. Given a fluorescence plant video taken over time, we first apply the alignment algorithm to the last frame of the video, and then continuously apply template transformation to the current leaf candidates in order to fit to the previous frame. We develop an objective function considering the Chamfer Matching distances, test image mask, and the rotation angels of all leaves. An example of multi-leaf tracking is shown below.
Back To Top