erned the 
is missing 
of output 
justments. 
ally was 
t had not 
n was the 
rk, which 
been able 
' cameras, 
of control 
rk would 
vould this 
ting large 
ould have 
tic errors. 
identified 
. network 
can yield 
antly, an 
ry cannot 
ys, poor. 
1 solution 
‘unctional 
/ bundles. 
, patient 
help the 
lisastrous 
mmercial 
JN 
liques to 
nterest to 
and the 
d further 
ecting a 
tionships 
d photo- 
bber was 
an array 
libration. 
itrol field 
distances 
ets were 
. Staff at 
e targets 
he video 
  
camera at a range of image scales. A single Panasonic 
NV-MC 10B CCD home video camera was used to 
obtain a series of images of the staff and control field 
from various positions, these images stored on TDK 
EHG EC30 video cassette. 
4.2 Image measurement 
Access to a PC based video frame grabber was 
provided and six images grabbed at a resolution of 748 
x 548 pixels in red/green/blue bands. Only the red 
images were used for further analysis. Image positions 
of the control and staff targets were measured using 
similar procedures to those discussed in Section 2.3. 
The geometric configuration between the six measured 
frames and the targets was convergent (Figure 6) and 
represented a network with stronger internal reliability 
than was possible in the betting office. 
Controlled tests. camera object relationships 
  
+f10 
i s 
Comera Positions 
| tf5 
pr 
I, S 
4 
par 
Pi 
81 state 
91 
61 
82 
93 
83 
64 Test field 
95 
5 metres 
Figure 6 
4.3 Data processing 
These measured data were processed using the self- 
calibrating bundle adjustment with and without the 
differential scale factor carried as an additional 
parameter. Staff object points were treated in an 
identical manner as foot and head points previously, 
with the Z ordinate held fixed for the staff reading of 
zero and the resulting XY coordinates held fixed for 
upper staff points. During the running of the self- 
calibrating bundle adjustments a problem of slow 
convergence was encountered, although only when 
differential scale factor was included. This problem 
was found to be associated with correlation between 
the exterior and inner orientation parameters and was 
particularly acute where the control field was planar. 
   
  
Four separate estimations were carried out in order to 
look at the role of the differential scale factor and the 
number of frames measured, either one or all six. An 
additional estimation was carried out using all 
measured frames, the optimum functional model and 
those taped distances and difference in height between 
the staff and the control points. This could be regarded 
as the best estimate for the staff position and staff 
height using all measured data. 
4.4 Results 
Table 1 indicates results obtained from these five 
different options. 
  
  
  
  
  
  
Table 1 
No. of Diff. Est. Plan Plan Var. Opt 
Frames Scale Staff position position factor 
Factor height of staff of staff a post. 
(1.600m) X) (Y) 
1 none 1.598 102.445 112.065 2.56 A 
+/-15mm 
6 none 1.568 102.499 112.183 2.00 B 
+/-2mm 
1 0.978 1.600 102.495 112.169 0.74 C 
+/15mm 
6 0.978 1.598 102.504 112.183 0.90 D 
+/-3mm 
64 0.978 1:591 102.506 112.195 0.87 E 
survey +/-3mm 
obs.