This “amateur” video camera (Fig. 3) was intentionally
used instead of an industrial CCD-camera, because of its
ease of operation (e.g. viewfinder for optimal object cov-
erage, internal compact video cassette for immediate data
storage of very long image sequences). Table 1 shows
some of the technical specifications of this camera.
Figure 3: JVC video camera GR-S77E
Table 1: Relevant technical specifications of the JVC
video camera GR-S77E
JVC video camera GR-S77E
Super VHS System for record and play mode
High resolution 1/2"-CCD-Chip (420 000 pixels)
Focal length 8.5 - 68 mm, 8x zoom lens
Auto focus
Variable electronic shutter 1/50, 1/250, 1/500, 1/1000 sec
Weight 1.2 kg
3.1. Image frame generation
The recording “flight” path of the video camera is illus-
trated in Figure 2a and 2b. The sequence of the 3-D test-
field was recorded in two strips, moving the “robot”
parallel to the testfield at an approximate distance of 3.6
meters from the wall. While recording the images, the
auto focus was switched off and the camera was focused
at infinity. With the focal length of the camera fixed at 8.5
mm, the depth-of-field can be assumed sufficient for sharp
imaging of the object. 53 seconds of the sequence have
been chosen for digitization. The imagery was digitized
with a VidcoPix framegrabber on a SPARCstation 1+
Figure 4: Three frames (no. 10, 15, and 20) of the vidco sequence (enhanced with a Wallis filter)
(Sun Microsystems). The generated image frames were
pre-processed with a low-pass filter (3 x 3 average). The
effective size of each digitized image was 720 (H) x 575
(V) pixels. Altogether 90 image frames were generated
giving a rate of 1.8 images per second of the sequence or
one digitized image every 0.6 seconds. Due to blurring ef-
fects caused by image motion, two image frames were left
out of the digitized sequence. Figure 4 shows three frames
of the complete sequence (enhanced with a Wallis filter).
The original visual quality of the frames is not very good.
Radiometric and thus geometric distortions due to motion
blur, analog video cassette storage, and frame grabbing
with PLL line-synchronization are visible if imaged at
larger scales.
32. Camera calibration
Before measuring image coordinates and processing data
in OLTRIS, the video camera was calibrated. In the cali-
bration, additional parameters including parameters of in-
terior orientation, x-scale factor, shear, and radial and
decentering distortion were determined. Investigations
into the calibration of CCD-cameras are described by
Beyer (1992). The respective software has also been used
here.
In addition to the test sequence, images were acquired for
calibrating the JVC. The 3-D testfield was imaged from
four different camera positions. The pixel coordinates of
the testfield targets were determined by least squares tem-
plate matching (LSTM), while reference coordinates for
the targets were obtained by theodolite measurements.
Measuring some well-distributed points in the four imag-
es yielded sufficiently precise approximations for the ex-
terior orientation of the four images by resection in space.
Using this data and the known object point coordinates,
approximate image coordinates could be computed. These
were used as initial values for automatic least squares
template matching of 130 points in each image. In this
test, LSTM was capable of measuring seven targets per
second including screen-display with an average precision
of 0.33 um (1/33 pixels) in x and 0.29 um (1/35 pixels) in
y.
The observations were processed in a bundle adjustment
with self-calibration. The measurements and adjustment
was performed in DEDIP (Development Environment for
Digital Photogrammetry, Beyer, 1987), which is a part of
the Digital Photogrammetric Station DIPS II (Gruen and
Beyer, 1990). The results from the bundle adjustment and
the comparison with check points, as an independent veri-
fication of the accuracy, are shown in Table 2. Version 1
Jade gqudilnub- JU NOM A ud Qd .f^AUuA Du MU // dh «ub Pun O5 uud m 44M BR o cu Ka [ÍÀÁ 4
