114
Table 3 Correlation matrix for the 12 parameters
Q $ K Xo Yo Zo Xo Vo f al az Di
1.00
o
$ |-0.38 1.00
K |-0.33 0.99 1.00
Xo| 0.33 -0.48 -0.47 1.00
Yo
Zo
0.69 -0.45 -0.45 0.18 1.00
-0.82 0.41 0.38 -0.20 -0.94 1.00
Xo|-0.16 0.91 0.92 -0.09 -0.34 0.25 1.00
yo|-0.99 0.34 0.28 -0.32 -0.60 0.78 0.12 1.00
f |-0.62 0.40 0.40 -0.14 -0.98 0.93 0.30 0.54 1.00
a1|-0.56 -0.01 -0.09 -0.19 -0.02 0.35 -0.20 0.67 0.02 1.00
a2|-0.39 0.58 0.60 -0.91 -0.39 0.37 0.24 0.35 0.37 0.09 1.00
pi| 0.61 -0.28 -0.21 0.25 0.12 -0.25 -0.14 -0.63 0.05 -0.43 -0.15 1.00
Table 3 shows a correlation matrix for the first image (Fig. 4) with 12 parameters. Strong correlation betweenw andy,
can be found and although strong correlation can be found also between some other parameters, these relationships
consist of more than two parameters, andthe value of w can be estimated as the zenith angle. Consequently, in
order to reduce interior parameters, let w be the known value. Since x and ÿ can be estimated as the horizontal
angle and 0 degrees respectively, let these values be the initial values because of the X,Y axis for the CCD camera and
the video theodolite are not perfectly parallel with the test course. :
Furthermore, since the information which is acquired from the goal and course lines in the orientation image give two
dimensional coordinates, let be 8 parameters, the exterior parameters (X,, Y,,Z,, « , ¢ ), principal point (x,,Y,) and focal
length (f) are unknown parameters in this investigation. In order to get stable solution, x, 256 was used
simultaneously as a additional constrain equation with collinearity equation.
Table 4 shows the calibration results and the accuracy for the orientation image with 16 control points.
Table 4 Calibration results and accuracy of a goal image
(a) Calibration results
E Y, Zs $ K x0 v0 f
0.258m -1.615m 0.650m -0°53'44"-0"6'3" 256.0 pic 218.2pc . 16.182
(b) R.M.S.E.
Xe Yo Lo $ K x0 vo f
105.460mm 17.074mm 6.253mm 0.049" 0.048? 0.233 pic 1.221 pic . 0.155mm
Figure 6 shows sequential images during the simulated 100 meter race. Since the doll was running on a parallel plane
perpendicular to the camera, X and Z coordinates for the doll's feature points (head and right-toe) in each sequential
image were calculated from the following two-dimensional projective transformation. The feature points were extracted
manually.
841X * àzy —asıf
xz fas 842X * a25y — aol (Vz Vo)
(1)
Z= Zu + FE yo)
842X * a25y — aol
where ;
Xo, Yo, Zo ; camera position
X , Y ,Z ; object coordinates of the feature points
X y ; image coordinates of the feature points
f ; focal length
aij; rotation matrix with three parameters,o,o,«
Figure 7 Multiexposure station Video theodolite
However, each sequential image was taken at a different exposure station as indicated in figure 7 dueto the discord
between the center of the theodolite and the lens of the CCD camera. Then each camera position have to be corrected to
response to the moving theodolite by following equation.
JAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences’, Zurich, March 22-24 1995