>> EPPN Website


Thomas Bodewein, Christoph Briese, Fabio Fiorani FZJ

Calibration procedures

Print the checkerboard pattern (Fig. 1) on a standard printer and fix (glue) it to a flat panel. Then, measure the size (x and y) of the squares to know their exact size.


Figure 1:  Checkerboard pattern

Size and dimension of the pattern can differ, as long as the same pattern is used for all images of one camera. For a good calibration, the number of squares along the x and y axis should be either even/odd or odd/even.

We recommend taking at least 20 pictures with different positions of the pattern in the images. Ensure that the pattern is also visible and covering the corners of the image frames to calculate possible lens distortion. The pattern has to be completely visible in every image. Fig.2 shows an example series of images used for calibration of the top-camera in the Screen House setup at IBG-2, Jülich.


Figure 2: Examples for different pattern positions, taken by the top camera

of Screen House imaging setup

For camera calibration, there are a number of software solutions available, e.g. Camera Calibration Toolbox for Matlab [1] or GML Camera Calibration Toolbox [2]. Another possibility is to write own software using e.g. the OpenCV library [3].

After loading the images, the calibration software should give you the following parameters for the camera:

  • Focal Length fc in pixel for x and y dimension
  • Principal point cc in pixel for x and y dimension
  • Skew coefficient alpha_c in degree
  • • Image distortion coefficents kc in a 5x1 vector

Note: Naming of the parameter can differ, according to the used software and literature. Based on the estimated parameter a distortion model for each camera can be calculated to cancel the effect of lens distortion in the images.

Image acquisition in a standard imaging cabinet” (Visser B.V, NL).

  •  The cabinet is illuminated by six 60 Watt halogen lamps. The imaging cabinet contains a motorized 360°-turntable and in three different positions a camera (Top-0°, Side-90°, Angle-45°).
  • The plants were positioned in the middle of the turntable and a 360° rotation was started.
  • The system was programmed to acquire one picture every 60°.
  • 6 pictures from different positions were recorded.
  •  Pictures from Top- and Side-Camera are analyzed (see “Screen House Image Processing Pipeline” in this document).
  • The “beginning position” of the plants on the turntable is changed in each of the three technical replicate series of the measurement.

Image segmentation and measured parameters

Image segmentation was done by using HSV-Segmentation as described in Walter et al 2007, [4]. Images were converted from RGB- to HSV color space and segmented by using a minimum and maximum threshold value for each single channel. Afterwards, the images have been cleaned up by closing holes in the segmentation and removing small objects (Fig. 4). This procedure is automated.

Based on the segmented images, the following parameter can be measured 3:

Projected Leaf Area: The number of pixels A belonging to the plant in the image, given by segmentation.

Plant Height: Height of the plant without pot in pixel or cm in images from side camera.

Contour Length: Length of the contour L from the plant in the image. A Pixel is considered as belonging to the contour if at least one of its neighbor pixels is designated as background. If there is another contour pixel in the 4-connected-neighbourhood of the pixel, the contour-length is increased by 1, otherwise by √2.

Area to Contour: Pixel Area divided by Contour length.

Gravity Center: Center of Gravity of the segmented pixel in the image.

Area Coverage: Relation of the plant area to the area of their convex hull.

Note that with the same data set additional geometrical parameters could be calculated, such as eccentricity, and also ‘biovolume’ [5].


Figure 3. Schematic representation of Screen House image processing pipeline. Note that color conversion is only needed, if the images are taken in raw- or bayer-pattern format. Undistortion is also only needed if there is a significant effect of lens distortion in the image data.


[1] Camera Calibration Toolbox for Matlab, http://www.vision.caltech.edu/bouguetj/calib_doc/

[2] GML Camera Calibration Toolbox, http://graphics.cs.msu.ru/en/node/909

[3] OpenCV, the Open Source Computer Vision Library, http://opencv.willowgarage.com/wiki/

[4] Walter et al, Dynamics of seedling growth acclimation towards altered light conditions can be quantified via GROWSCREEN: a setup and procedure designed for rapid optical phenotyping of different plant species, New Phytologist, 2007

[5] http://www.lemnatec.com/sites/default/files/application-sheets/2009/12/27/66_Influence_of_alignment_on_biovolume.pdf