This title appears in the Scientific Report :
2010
Please use the identifier:
http://hdl.handle.net/2128/3725 in citations.
Plant Leaf Motion Estimation Using A 5D Affine Optical Flow Model
Plant Leaf Motion Estimation Using A 5D Affine Optical Flow Model
High accuracy motion analysis of plant leafs is of great interest for plant physiology, e.g., estimation of plant leaf orientation, or temporal and spatial growth maps, which are determined by divergence of 3D leaf motion. In this work a new method for plant leaf motion estimation is presented. The...
Saved in:
Personal Name(s): | Schuchert, T. |
---|---|
Contributing Institute: |
Phytosphäre; ICG-3 |
Imprint: |
2010
|
Dissertation Note: |
Aachen, RWTH, Diss., 2010 |
Document Type: |
Dissertation / PhD Thesis |
Research Program: |
Terrestrische Umwelt |
Subject (ZB): | |
Link: |
OpenAccess |
Publikationsportal JuSER |
High accuracy motion analysis of plant leafs is of great interest for plant physiology,
e.g., estimation of plant leaf orientation, or temporal and spatial growth maps, which
are determined by divergence of 3D leaf motion. In this work a new method for plant
leaf motion estimation is presented.
The model is based on 5D affine optical flow,
which allows simultaneous estimation of 3D structure, normals and 3D motion of objects using multi camera data.
The method consists of several consecutive estimation procedures. In a first step the affine transformation in a 5D data set, i.e., 3D image sequences (x,y,t) of a 2D
camera grid (sx,sy) is estimated within a differential framework. In this work the
differential framework, based on an optical flow model, is extended by explicitly
modeling of illumination changes.
A second estimation process yields 3D structure and 3D motion parameters from
the affine optical flow parameters. Modeling the 3D scene with local surface patches
allows to derive a matrix defining the projection of 3D structure and 3D motion onto
each camera sensor. The inverse projection matrix is used to estimate 3D structure
(depth and surface normals) and 3D motion, including translation, rotation and
acceleration from up to 24 affine optical flow parameters.
In order to stabilize the estimation process optical flow parameters are estimated
additionally separated for all cameras. A least squares estimator yields the solution
minimizing the difference between optical flow parameters and the back projection
of the 3D scene motion onto all cameras.
Experiments on synthetic data demonstrate improved accuracy and improved robustness
against illumination changes compared to methods proposed in recent literature. Moreover the new method allows estimation of additional parameters like surface normals, rotation and acceleration. Finally, plant data acquired under typical laboratory conditions is analyzed, showing the applicability of the method for plant physiology. |