“the 
etric 
uate 
Is to 
der 
em 
ned 
the 
ion 
> as 
| to 
ve) 
ent 
of 
er, 
cal 
on 
his 
no 
ip 
on 
on 
es 
he 
he 
er 
ait 
Carter, John 
  
links the measured and the real gait angle. As in Eqn. 1, a simple scaling is suggested, but since the horizontal 
component of gait angle carries the majority of the angular information this cannot be accepted without proof. 
  
Figure 3 Simulated legs, hip rotation angle 12.2 , 
trajectory angle 14 and ~2m from the camera. 
Uncorrected Angle Data 
  
T T T T 
Un c 
e o 
47 T 
1 
# 
40 * 
— N WW 
eoo o 
P 
Measured Angle (Degrees) 
  
4d SW GER 0n a 
4448 446€ © € 
ÿ 8 44839600 9 
4d og de«og sx 
1 
Le 
L 
0 
Trajectory Angle (Degrees) 
  
  
$ 3 ad dd Gà à à 
a8 48 8083 9 93 
© 
— 
© 
N 
© 
UJ 
A 
© 
Un 
© 
Figure 4 Measure angle, as a function of 
trajectory angle for camera/target distances 
between 1 and 4 meters. Diamonds are closest to 
the camera. 
Corrected Angle Data 
  
  
  
  
~~ 
o 
D 
D T T T T 
C» 
2 60 + 7 
NN DO: v 9 9 v 9 | 
© 40% = =» 8 8 8 9 | 
< 30 9$. 9 m GB 8: $. | 
wo d" wv v.v a ws 9 
D 20 &  €9 9 ww S. 9. il 
E we wv * = = *— X 
0 10 8g v wv 9 S $$. 
e B oU v" 9" © e wu 
= 0 ala ls | al 
0: 10-20 :30- 40-50 
Trajectory Angle (Degrees) 
Figure 6 Measured angle, after application of 
trajectory correction to each half angle. 
3.3 Laboratory Experiments 
First, the hip rotation angle was simulated, by printing a target 
pattern as shown in figure 3. A walking subject was simulated 
by a printed set of black circles, representing the hip and two 
knee positions. A set of seven targets representing different 
angles from zero to 45 degrees was used. Each target was 
mounted, in turn, on a rotating table and viewed with an 
uncalibrated camera from a fixed distance. A Sony Video 
camera model XC-711P with a Sony Zoom lens Model TV- 
ZOOM (12.5mm to 75mm) was used for this and all other 
experiments. The zoom lens was adjusted such that the target 
filled as much of the field of view as possible. The output from 
the camera was digitised and a global threshold was applied to 
produce a binary image. After the circle positions had been 
marked manually, a minimum sized bounding box was 
generated automatically, and the centre of gravity of the circle 
was calculated. For each group of circles the apparent hip angle 
(y) calculated and recorded as a function of simulated trajectory 
angle. Figure 4 shows the raw measured angle, for trajectory 
angles between 0° and 45°, for viewing distances of between 1 
and 4 metres. Clearly, increasing the trajectory angle results in a 
decrease in the perceived angle. 
Figure 5 shows the corrected hip rotation angles after the 
correction in Equation (2) has been applied directly. As can be 
seen, this naive approach is not effective, as the perceived angle 
Naively Corrected Angle Data 
  
  
  
  
^ 
D 
o T T T T 
C 
2 60 r j 
eed 50 ¢ ® © | 
o 291807 
o 40 = © e g g g g _ 
€ 3015454? ,89. 9 9 w 
= "S 5 ww +% v v 
D 20 Ÿ © v = u = & | 
5 Y = = = = wow 
® 40% © v 9v © *o € | 
© 
o y "iS 9. % w = 
S 0$9—9-——9—3-——9— 9 —e 
0" 10. 20 30 40*' 50 
Trajectory Angle (Degrees) 
Figure 5 Application of Eqn. 2, to the data 
shown in Figure 4. Note the remaining 
dependence on trajectory angle. 
is not independent of the trajectory angle, unlike figure 6. In 
figure 6, each half angle is considered separately and the 
horizontal components are corrected before the angles are 
calculated. The measured angles are now correct for all 
trajectory angles less than 45 and at distances greater than 1 
meter. Detailed examination of the data reveals that there is 
some inaccuracy at large angles and when the camera is close 
to the target, again this is to be expected. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 117