68 
Fishery Bulletin 99(1 ) 
0.0 
- 0.2 ' 
-0.4 
E -°- 6 1 
(_> 
o 
lu "0.8 
- 1.0 | 
- 1.2 1 
0% T 
o 
> 
o 
> 
- 1 % 
2 - 2 % 
a5 -3% 
cc 
-4% 1 
-5% 
a .. 
1.40 
1.20 t 
! 
_ 1.00 
E 
0 0.80 $ 
cd 
CD 
1 0.60 - 
o 
c/) 
n 
< 0.40 | 
0.20 i 
0.00 
5% y 
4% | 
o 
CD 
§ 3% 
<D 
It 
CD 
Cl 1 
Jill 
> 
o 
> 
1 2 % 
1% I- 
0% 
Figure 4 
Means and 95% confidence intervals for each measure of error, for 
each stereo-video operator. 
9% | 
8 % 
7% -- 
6% 
5% + 
4% 
3% 
2 % + 
1% 
0% 
O 
o 
+ Good 
O Poor 
+ 
<*> % 
o 
+ 
+ 
o<^ 
0 
2 3 
True length (cm) 
Figure 5 
Relative absolute error for stereo-video estimates of 
silhouette length under good and poor water clarity. 
were 18.2%, 18.6%, and 1.6% for novice scientific 
divers, experienced scientific divers, and the ste- 
reo-video system, respectively. The results of the 
experienced divers’ estimates are similar to those 
obtained in other published research studies (Bell 
et ah, 1985). 
In general, water clarity does not appear to 
affect the accuracy or precision of the measure- 
ments made from the stereo-video system. The ac- 
curacy of the of the stereo-video system is limited 
to the ability of the operator to accurately point 
out image locations of interest that are then re- 
corded to subpixel resolution. Discrete sampling 
of the CCD (charged coupled device) sensors, com- 
bined with noise artifacts from the video tape re- 
cording and frame grabbing, tends to smear the 
edges of the images and blur details (Shortis et 
al., 1993). This result is particularly noticeable 
under recording conditions with good water clar- 
ity and high contrast where there is a problem 
with the detection of the edge of the silhouette. 
The edges or outline of the points of interest (in 
this instance, the snout and fork of the tail of each 
silhouette) become significantly blurred within 
the computer image owing to sampling and noise 
effects on each pixel. There is a tendency for the 
observer to select a location inside the true edge 
of the point of interest because the location most 
nearly matches the local appearance of the body 
of the object of interest. This results in the under- 
estimation of silhouette length, as demonstrated 
in Figure 1. Under poor water clarity, the edges of 
the silhouettes have less contrast with the dark 
background. This lack of contrast permits more 
accurate pointing to the edges of the object of in- 
terest because the sampling and noise effects gen- 
erate a smaller disparity between the real and 
apparent edges (Fig. 8). Less accurate measure- 
ments will be made when the contrast becomes 
so low that the operator cannot discriminate be- 
tween the object of interest and the background. Hence, 
where there is sufficient contrast to discriminate the ob- 
ject from the background, lower contrast will realize more 
accurate values that ameliorate the underestimation of 
length. 
Advantages of stereo-video census techniques 
The use of a stereo-video system for the measurement 
of reef-fish length has many advantages. It significantly 
decreases measurement error and is relatively insensitive 
to operator experience. The data suggest that the stereo- 
video system provides far greater accuracy and precision 
than even experienced scientific divers, and potentially 
allows inexperienced volunteers to participate in monitor- 
ing programs without compromising the accuracy or pre- 
cision of the data collected. This degree of accuracy and 
precision may be important in research where the objec- 
tive is to detect small (5-30%) changes in the mean length 
of a population or assemblage of reef fish, with a high level 
