1071
the camera position and orientation in regards to the 3-D scene (rotation and translation from the 3-D origin).
More details can be found in Toscani (1988) and an example of the calibration in field conditions has been
published in Ivanov et al. (1994).
The explored vegetation canopy was ZEA mays L., DEA cultivar with density 0.8 m between rows and
0.15 m along rows. The experiment took place in the fields of I.N.R.A. Grignon, France (30 km west from
Paris), in the beginning of August 1993, when maize reached its maximum height of 2.5 m after tasselling.
Images of 100 maize plants in 4 rows (figure 2) were acquired by two photographic cameras (6 x 6 cm film and
100 mm focal length) located at H = 8.5 m above the ground by using a mechanical arm . The optical axes were
slightly converging to obtain a 100 % overlapping of the 4 x 4 m 2 area covered by the two cameras. Distance
between the cameras, called "base B", was equal to 1.05 m which gave a ratio B / H = 0.12. This ratio is an
important parameter that determines the precision of the estimates as it is showed further in the paper. The
cameras were remotely controlled by a synchronising device allowing simultaneous release. Photographs were
taken on Kodak Ectachrome colour slides and enlarged on 18 x 18 cm paper prints.
To enable the reconstruction of the full maize canopy, a destructive sequential procedure has been
performed. It consisted in successively cutting leaf layers, starting from the top of the plants, and taking a pair
of images each time when a leaf layer has been eliminated. Thus, we obtained a set of 11 stereo photographs
where all leaves and stems could be easily identified. The photographs were processed in the reverse order than
they have been taken. The image pair, where only the stems were present, was treated first. Then the following
pair, with only the most lower down leaf layer, was processed. In this way, all photographs were successively
operated from the bottom to the top of the vegetation cover, taking only one leaf layer per photograph.
2.3. Stereo Matching
The stereo matching procedure includes two main levels. The first one concerns the identification and
extraction of leaf contours from the images. Once a leaf from the left image was recognised as the same leaf in
the right image, they were both identified by the same numeric label and that represented the preliminary stage
of matching. Such leaves were called "homologous" leaves. The contour extraction was performed manually
using a 2-D digitizer. Starting from the tip and going to the insertion point on stem of a given leaf, the operator
followed its edge with a pointer, the position of which was recorded with 0.2 mm precision. After the manual
boundary extraction, the edges contained points unequally distanced along the contour. Therefore, it was
necessary to interpolate them according to a 2-D grid in order to obtain regularly spaced points along
horizontal and vertical directions with a given step resolution (figure 3). contour (Chapron et al. 1992). Then
the second matching level was carried out point by point, starting from the tip and going to the insertion point
of the homologous leaves so as to create pairs of homologous points.
