Walker et al.: Use of an underwater camera to monitor distribution and density of Placopecten magellanicus 
265 
Distortion contours (measured from 1 to 7 pixels) in rela- 
tion to the center of an image for the AUVs underwater 
camera used in a study of the distribution and abundance 
of the sea scallop {Placopecten magellanicus) during 2011. 
and used to correct the calculation, although in other 
studies (Gudmundsson, 2012; Singh et ah, 2013, 2014) 
both pitch and roll were found to be negligible factors. 
Not accounting for pitch would have resulted in a po- 
tential 3% (mean) overestimation of image height for 
all of the photos. For Equation 1, the AUV is assumed 
to image a flat seafloor over the area of the full frame, 
and the equation also does not account for roll of the 
vehicle. The roll-induced error associated with image 
width is less than 1.0% (at 2 m altitude) if vehicle roll 
is less than 10° from horizontal, and log data showed 
that the AUV operated with roll characteristics of 
x=4.01°C and cf=1.11° for all survey sites. Singh et al. 
(2013 and 2014) reported a similar ground distance er- 
ror (<2%) due to both AUV pitch and roll for the same 
Scorpion 20SO camera when surveying at an altitude 
of 2 m. Similarly, Gudmundsson (2012) performed as 
detailed a calibration of the same AUV camera system 
as that used in our study and reported negligible ef- 
fects of camera distortion, pitch, and roll. 
Scallop counts and sizing 
between 0.23 and 0.46 cm in distance on the ground. 
This distortion uncertainty is approximately within 
<5% of the average shell height directly measured in 
the dredge tow samples, with which the image-based 
measurements were favorably compared. It is impor- 
tant to note, however, that camera distortions have no 
impact on the enumeration of scallops and the result- 
ing analysis of scallop counts. In previously published 
studies of sea scallop shell height and abundance, this 
same combination of AUV camera was used and cali- 
brated camera distortions along with both pitch and 
roll of the AUV were found to be negligible (GuSmunds- 
son, 2012; Singh et al. 2013; Singh et al., 2014). Our 
study, therefore, is consistent with the findings of the 
previous research cited above, suggesting that the in- 
fluence of roll bias (< 1%), camera distortions (< 1%), 
and manual digitization (< 1%) overall contributes less 
than 5% uncertainty for estimates of shell height. The 
width (W) of a single image was determined by using 
the image metadata collected by the AUV navigational 
system and sensors. 
W = 2tan 
[z-1.3siii(-0 )], 
(Eq. 1) 
where W = 
a,, = 
2 = 
0p = 
seafloor image width in meters; 
horizontal viewing angle (degrees) of the 
camera in water; 
height above the seafloor in meters; and 
pitch of the AUV in degrees. 
Knowing the horizontal viewing angle of the cam- 
era in water (a;,=44,65°), and the height above the sea- 
floor iz), we were able to calculate real world dimen- 
sions on the seafloor in each image. The pitch of the 
AUV (Op) and the arm length (1.3 m) from the camera 
to the AUV navigational reference point were known 
The 22 AUV surveys resulted in 203,066 images of 
the seafloor; see Figure 3 for a selection of represen- 
tative images. In order to process all of the images, 
we engaged a team of graduate students and interns 
to count and size scallops using software written in- 
house for this project. Each scallop annotator received 
training on identifying sea scallops in benthic images, 
and was required to successfully identify at least 95% 
of the scallops from a standardized image data set be- 
fore being allowed to annotate the rest of the images. 
Repeated digital measurement of the same scallop by 
the same annotator (N=53, where N is the number of 
sea scallops) yielded a standard deviation of 5.0 mm 
in shell height measurement. This value of annotation 
repeatability for size determination is in agreement 
with the 5 mm annotator measurement error reported 
by Singh et al. (2014). Furthermore, it is worth noting 
that in comparison with manually sized scallops from 
dredge samples, manually sizing was itself segmented 
into 5-mm bin intervals and thus the discretization of 
image-based sizes was on par with the discretization 
from physical samples. The protocol used for image se- 
lection and sizing was the following: 
1 All images that were taken at a height between 1.5 
m and 2.5 m above the seafloor were counted (re- 
moving the starting descent and ending ascent por- 
tions of each survey). 
2 Each sea scallop in an image was counted individu- 
ally, unless it had already been counted from the 
previous image that overlapped the same section of 
seafloor. Annotators examined photos sequentially 
and were trained to recognize overlapped images, so 
that scallops were not counted more than once. 
3 Each scallop shell height was sized on the basis of 
the distance from the shell umbo to the ventral mar- 
gin by using a pixel-measuring tool. The projected 
