Content uploaded by Christopher N. Rooper
Author content
All content in this area was uploaded by Christopher N. Rooper on May 16, 2015
Content may be subject to copyright.
352
For moder n f i sher ies sto ck assess-
ments, fisheries-i ndep endent data
are necessary to estimate population
abu ndances and population trends.
For most marine species, f isheries-
independent abundance estimates are
primarily obtained from large -scale
multispecies bot tom trawl survey s
(e.g., Gunderson a nd Sample, 1980)
and from acoustic surveys of pelagic
fish stocks (e.g., Karp and Walters,
1994). Although acoustic backscatter
is used to measure fish abundance,
midwater trawl samples are needed
to det ermi ne the si ze and sp ecies
composition of acoustically sampled
fish populations. Both of these sur vey
methods requi re physical sampling
of trawl catches and such sampling
can result in unrepresentative data
in several ways.
Bottom-trawl surveys a re limit ed
to th e area s they ca n sa mple be -
cause many research trawls are not
constructed to ef ficiently fish over a
rough or rugose seafloor. Thus, sur-
veys with bottom trawls may not be
appropriate for some species with
affinities for untrawlable habitat or
in survey areas where sig n ific ant
patches of untrawlable ground can be
found (Zimmerman n, 2003; Cordue,
2007). In Alaska, semipelagic species
such as northern rockfish (Sebastes
polyspinis) and Pacific ocean perch (S.
alutus) are an important part of the
commercial catch, but they also show
some affinity for untrawlable areas
Use of stereo camera systems
for assessment of rockfish abundance
in untrawlable areas and for recording
pollock behavior during midwater trawls
Kresimir Williams (contact author)
Christopher N. Rooper
Rick Towler
Email address for contact author: Kresimir.Williams@noaa.gov
Alaska Fisheries Science Center
National Marine Fisheries Ser vice
National Oceanic and Atmospheric Administration
7600 Sand Point Way NE
Seattle, Washington 98115
Manuscr ipt submitted 21 January 2010.
Manuscr ipt accepted 27 May 2010.
Fish. Bull. 108:352–362 (2010).
The views and opinions expressed
or implied in this article are those of the
author (or authors) and do not necessarily
reflect the position of the National Marine
Fisheries Service, NOAA.
Abst rac t—We des cr ibe the applica-
tion of two typ es of ster eo camera
systems in fisheries research, includ-
ing the de sig n, c al ibration, analysi s
techniques, and precision of the data
obtained with these systems. The first
is a stereo v ide o system deployed by
using a quick-responding winch with a
live feed to provide species- and size -
composition data adequate to produce
acoustically based biomass estimates
of rockfish. T his system was tested on
the easter n Bering Sea slope where
ro ck f ish were mea su red . Rockf ish
sizes were similar to those sampled
with a bottom trawl and the r elative
error in multiple measurements of the
same rockf ish in multiple stil l-frame
im a ges wa s s mal l . M eas ure ment
err ors of up to 5.5% were fou nd on a
cal ibration target of k nown size. The
sec ond syst em consist ed of a pair of
sti ll-image digit al ca meras mounted
inside a m idwater t rawl. Processing
of the st er eo ima ge s al lowed f is h
len gth, fis h orient at ion in relation
to the camera plat fo rm, and rel a-
tive d ist ance of the fish to the trawl
netting to be determined. The v ideo
system was useful for sur vey ing f ish
in Alaska, but it could also be used
broadly in other situat ions where it
is diff icult to obtain species-compo -
sit ion or size -compo sitio n infor ma -
tion. Likewise, the still-image system
cou ld be used for fisheries research
to obtain data on size, po sition, and
orientation of fish.
(Clausen and Heifetz, 2002; R ooper
et al., 2007).
In add ition, inferences from spe-
cies- and size- composition data ob-
ta i ned from trawl ca tch es can be
biased on account of trawl selectiv-
ity. Trawls are gener ally des igned
to capture larger, market-size fish,
an d their desi g n for this select ed
size results in the under-retention
of juven ile size classes. In acoustic
surveys of walleye pollock (Theragra
chalcogramma), biases in midwater
trawl catches directly translate into
biases in abunda nce estim at es for
areas where large and small fish are
found (Godo et al., 1998). Selective
retention of fish is a consequence of
size and spec ies-dependent fish be-
havior du ring the trawling process.
Observation of fish reactions to trawl
gear is critical to understanding the
behavioral mecha nisms resp onsible
for t rawl selectivity and to develop
future trawl gear for research.
Here, we describe the use of stereo
photography to sample rockfish in un-
trawlable habitats using a drop unit
with a stereo video camera (hereaf-
ter termed “video-drop” camera), and
to study fish behavior i n midwater
trawls using a trawl-mounted pair of
still-frame stereo cameras (hereafter,
termed “still-frame” camera). Stereo
cameras have been successfully used
to measure fish in controlled aquacul-
ture settings (Ruff et al., 1995 ; Har-
vey et al., 2003) and in open water
353
Williams et al.: Use of ster eo-camera systems i n assessing roc kfish abundance and pollock behavior
(i.e., van Rooij and Videler, 1996; Shortis et al., 2009).
The recent development of high-resolution digital cam-
eras has vastly improved the performance and reduced
the complexity of image-based sampling because high-
quality digital images can be directly analyzed with
image-processing software. In general, stereo methods
provide highly precise measurements in comparison to
single-camera–based photogrammetric methods (Har-
vey et al., 2002). However, these systems necessitate
maintaining a stable two-camera geometry and must be
initially calibrated with targets of known sizes. Despite
these constraints, stereo photography is widely used in
optical-based sampling in a variety of marine studies.
We demonstrate the precision of stereo-camera–based
measurements, attainable from initial deployments in
the field, in comparison with traditional survey mea-
surements. The results show that stereo-based optical
sampling is a viable method for augmenting bottom-
trawl data for abundance estimations; the stereo cam-
eras allow scientists to survey sampling areas that are
unavailable to standard survey trawl gear. In addition,
stereo cameras can be used t o observe a nd quantify
the behavior of fish in the process of being captured by
trawl gear to further improve estimates of abundance
Table 1
Design, manufacturer, and cost (approximate estimates in U.S. dollars) for drop stereo-video camera and still-frame stereo-
camera systems used for surveying untrawlable habitat and studying fish behavior in midwater research trawls. Both sys-
tems were used in the field in July 2008 and July 2007, respectively. HID =high-intensity discharge; LED=light-emitting diode;
UHMW = ultra high molecular weight plastic.
System Component Design Manufacturer Cost
Drop stereo- HID light HID Xenon lights, 12 V, 50 W Underwater Lights USA $814
video Video line driver Balanced line driver and transceiver Nitek $133
camera Conducting cable 4 conductor wire, 4.72 mm diameter Rochester Cable $1601
Sled frame Aluminum channel and tubing Local manufacture $2000
Winch and slip ring CSW-6 electronic win A.G.O. Environmental $11,268
Underwater housings 5ʺ diameter Local manufacture $729
cameras
Underwater housings — Local manufacture $729
lights
LED sync — Ramsey Electronics $24
Underwater cable and — Teledyne Impulse $614
connections
Batteries 4 × 12 V 4 Ah NiMH Energy sales $396
Total video system cost $18,308
Still-frame Strobe Oceanic 3000 Oceanic $990
stereo Cameras Canon Digital Rebel Xt (8Mp) Canon USA $1100
camera Lenses Canon EF 28 mm f /2.8 Canon USA $450
Microcontroller & — Local manufacture $150
circuitry
Underwater housings 10ʺ floats, 1.5ʺ acrylic flat viewports Local manufacture $1400
and viewports
Mounting frame UHMW plastic and aluminium stock Local manufacture $350
Underwater connections — Teledyne Impulse $650
Batteries 3 × 12 V 4 Ah NiMH Energy sales $297
Total still-frame system cost $ 5387
Software Matlab V 7.6 Mathworks
because they allow scientists to determine the potential
biases in trawl-based catch data.
Materials and methods
Sampling untrawlable are as with the
video -drop c amera syst em
The design of the video-drop system was based on two
key needs. Because rockfish are found in areas of high
relief, the camera needed to have adequate protection
for their electronic components and have the ability to
maintain visual contact with the bottom through rough
substrate areas. Therefore, essential to sampling with
this camera system was the ability to live-view the video
and the use of a quick-responding winch system that
could be controlled by the operator aboard the research
vessel. The specifications of the camera components are
presented in Table 1.
The w inch used t o deploy and retrieve the camera
system and navigate the seafloor was a CSW-6 multi-
purp ose w in (A.G.O. Environ mental, Nanaimo, BC,
Canada; Fig. 1 A). The winch motor was a 3/4 horse-
354 Fishery Bulletin 108(3)
A B
CDD
winch cameras
cameras
strobe
still
camera
lights
Figure 1
Drop stereo-video camera system showing (A) the quick-responding winch, and (B) the locations of
cameras and lights i n the a luminum fra me. (C) The sti ll-frame camera system, which is deployed
in a m idwater trawl. (D) The camera housing s (shown) were con struc ted from deep-water trawl
floats which provided buoya ncy and reduc ed the weight of the cameras.
power Leeson wet-duty motor powered by 100 V AC.
The winch speed ranged from 43 m /min (with a bare
drum) to approximately 58 m/min (with a full drum).
The approximate square-in area of the winch was 48
in 2 (0.031 m2) and its weight was 155 lb (70. 3 kg).
The drum was 16 in (40.6 cm) in circum ference and
was filled w ith 1312 ft (40 0 m) of 3/16 -in (4.72-mm)
conducting wire. The wire had a breaking strength of
3300 lb (1497 kg) and was connected to the camera sled
with a cable-grip. The video feed from the cameras was
passed up a cable and through a fou r- conductor slip
ring mounted on the winch and routed into a junction
box where it was connected to a monitor for real-time
viewing.
The protective cage around the camera and lights was
constructed of 1.5-in (3.81-cm) aluminum tubing, and
the interior members of the frame were composed of 6-
in (15.24-cm) aluminum channel (Fig. 1B). A tail chain
was attached to the rear of the ventral surface of the
cage to drag along the seafloor to help keep the camera
unit in contact with the seafloor and oriented forward
during deployment. The tail chain was connected to the
cage by a short piece of twine to act as a “weak link” in
case the tail chain snagged on the seafloor.
The underwater video was recorded by two identical
Sony T R-900 camcorders (Sony Electronics Inc., San
Diego, CA) located inside the camera housi ng s. The
cameras were capable of collecting 720 p progressive
scan video images at a resolution of 720
× 480 pixels.
The video was recorded to digita l v ideo tapes for a
max imum of one hour per tape. Because the cable was
too long (400 m) to transmit a standard video signal, it
was transformed by using a video balun (in the camera
housing) and a receiver (at the winch) to reconvert the
video signa l back to a v iewable picture of the seaf loor
to use for real-time navigation. Cameras were placed
in separat e housings constructed of titanium tubing
and that had a gla ss dome por t (pr essure -r at ed to
30 0 0 m depth) cover i ng the lens. The lens of eac h
camera was keyed to its port t o prevent the camera
fr om bei ng insert ed into the hou sing in a p osit ion
other than the exact keyed position and stabilized the
relative position of the camera s from deployment to
deploy ment —an important consideration for accurate
355
Williams et al.: Use of ster eo-camera systems i n assessing roc kfish abundance and pollock behavior
measurement of target s ( Short is et al., 20 00). The
housings were mounted side by side on the aluminum
frame (Fig. 1B).
Illumi nation was provided by two lights mounted
above the camera housings inside the aluminum frame
(Fig. 1B). The lights were 50-watt high-intensity dis-
charge ( HID ) Xenon lig hts with 5300 lumen output
and 3900 Kelvin color temperature. The lights were
inserted into 3-in (7.62- cm) dia met er titaniu m hous-
ings and the entire light weighed 5 lb (2.27 kg). The
lights were powered by a battery located in the camera
housing and linked to the light housing by under water
connectors. Four rechargeable 4 A h 12 V nickel-metal
hydride batteries were connected in parallel to provide
approximately 1.5 hour of light per deployment. Each
light housing was mounted on an adjustable mount that
allowed even illumination of the target.
Obser ving fish behavior in a trawl
wit h t he still-frame system
The still-frame system was designed to be light and
small enough to be easily attached to the inside of a
survey trawl without significantly changing the fishing
activity of the net. The system also needed to provide
adequate illumination and resolution in order to allow
the fish inside the net to be observed at a range of up
to 6 m as they passed though a midwater survey trawl
40 m ahead of the codend. A pair of Canon Rebel Xt
8 megapixel digital single-lens reflex cameras (Canon
USA, Lake Success, NY ) were used to capture fish
images. Both cameras were outfitted with 4-gigabyte
compact flash memory cards for storage of the images.
A high-power wide-angle Xenon strobe (90°, 150 W/s)
was used to illuminate the field of view. Three 4-Ah 12
V batteries were mount ed in the strobe housing; two
were connected to the strobes and the third was used
to power the cameras.
The camer as were mounted in sepa rate hous ings
made from 10-in (25-cm) diameter deep-water–rated
(1800 m) trawl floats. Images were taken though a 25-
mm thick flat acrylic viewport. The strobe and batteries
were mounted in a thi rd float housing (Fig. 1C). A ll
three float housings were secured on a sled constructed
of 25-m m thick plastic plate and alum inum rails for
protection. The approximate weight of the complete as-
sembly in air was 30 kg and was positively buoyant be-
cause of the float housings. Quick-release trigger snaps
were attached to the ends of the plastic mounting board
for attachment to the inside of the trawl. The cameras
were aimed across the trawl, perpendicular to the wa-
ter flow to provide lateral views of fish passing by. The
trigger on the camera shutter was controlled by using a
microprocessor that was programmed for the study and
that located in one of the camera housings. A two-axis
tilt sensor was attached to the microprocessor board to
allow measurements of fish tilt (deviation of snout-tail
axis from the horizontal) and yaw (angle of fish head-
ing in the horizontal plane) to be adjusted from being
relative to the camera platform to being in absolute
orientation. A pressure switch was used to activate the
system once the depth exceeded 20 meters. Images were
taken at intervals of 5 s to reduce the influence of light
on fish behavior and to ensure that a new group of fish
was observed in each frame. The system was capable
of taking about 400 images or operating for 33 min of
trawl time per deployment.
Calibration of the two t yp es of stereo cameras
The same calibration procedure was used for both stereo-
camera systems. The basic procedure required collecting
images of a target plate with a pr inted 10
× 10 square
checkerboard pattern of known dimensions (50
× 50 cm
squares for the video-drop system, 100
× 100 cm for the
still-frame syst em). T his calibration was performed
underwater. The video-drop system cage was suspended
in the water while the research vessel was secured to
the dock. The approximate depth of the camera was
1 m and the distance from the target was 2 m. The
checkerboard target was lowered into the water along
the vessel until it was plainly visible in both cameras.
The target was then slowly moved horizonta lly and
vertically through the field of v iew of both cameras.
Up to 15 min of calibration video was collected by this
method. For the still-frame system, an external trigger
cable was attached to the assembly, a nd t he system
slowly moved about while capturing images of the fixed
checkerboard plate.
To calibrate the video-drop system, progressive scan
video images were collected at 29.97 frames /s in each
camera, and the beg inning of the v ideo feed from each
camera was aligned by using a light-emitting diode
(LED) synchronization light at the beg inn ing of de-
ployment. This process was repeated at the end of the
deployment to confirm that the video frames were still
aligned. For the calibration procedure, still frame im-
ages were extracted from the aligned video at 1-s in-
tervals with Adobe Premier software (Adobe Systems,
Inc., San Jose, CA). Synchronization was not necessary
for the still-f rame system because the cameras were
triggered simultaneously. Approximately 20 paired im-
ages where the target checkerboard was visible in both
cameras were randomly selected for the calibration of
each camera system.
The calibration parameters were estimated with the
camera calibration toolbox in Matlab, a freely available
software analysis toolbox built with Matlab computing
lang uage (Mathworks, Inc.; Bouget, 2008; Fig. 2). For
each image pair, the position of the corner points of the
checkerboard pattern were identified by clicking on the
images and the location of these points in the still im-
ages was computed by the calibration software to deter-
mine the intrinsic parameters of each camera. Intrinsic
parameters were used to correct the individual images
for optical distortion resulting from the camera lenses.
The checkerboard pattern allowed the software to auto-
matically pinpoint exact corner locations based on the
color contrast of the squa re boundaries, making the
initial precision of the manual clicking less critical.
356 Fishery Bulletin 108(3)
1 m
1 m
Figure 2
The checkerboard patt ern used to calibrate and
test the drop stereo-video ca mera and the still-
fra me st ere o- ca mera systems. Image s use d for
cal ibration were pro cessed by u sin g the camera
calibration toolbox written for Matlab. A n image
taken by t he stil l-f rame st ereo- ca mera system
is shown with the user-selected extreme corners
shown as white circles and the automated extrac-
tion of all intermediate corners shown with white
cro sses. This proc edu re was repeated for up to
20 different image s with both systems.
Stereo calibration required that the checkerboard
corners be identified in the same order in each of the
synchronous image pair s to c or rectly match up the
analogous corner points. These points, once corrected
for optical distortion in individual cameras, were used
to compute the epipolar geometry, by iteratively solving
for the translation and rotation vectors that describe
the relationship between the coordinate systems of the
two cameras (Xu and Zhang, 1996). Once these matrices
were estimated by the software, the three-dimensional
position of a target point viewed in both cameras could
be determined by triangulation.
Fis h measur ements with the came ra
Fish lengths were measured by using stereo triangu-
lation functions supplied w ith the camera calibration
software package (Bouguet, 2008). For the video-drop
system, images were extracted from the two video feeds
at 1-s intervals. The images were synchronized at the
beginning of each transect before deploy ment by using
the LED synchronization light. The images were checked
at the end of each transect to confirm that the cameras
remained synchronized.
Length measurements were obta ined by identifying
the pixel coord inates of corresponding pixel locations
in the left and r ight camera still frames such as a
fish snout and tail (Fig. 3). T hese points were used
to solve for the three-dimensiona l coordinates of the
poi nts in the images by t riang ulation, by using the
ca libration-derived para meters. Once the three-d i-
mensional coordinates of the fish snout and tail were
obta ined, the leng th was measured as the simple Eu-
clidian distance between the points in real space. This
measurement method underestimated leng th for fish
whose bodies were curved; however; f ish in the video
and still camera were a lmost exclusively seen with
little or no curvature in their bod ies and the few indi-
viduals that were obviously strongly curved were not
measured. Length data were collected by using a basic
software application bu ilt with the Matlab computing
language (Fig. 4; available f rom the authors upon re-
quest), which incorporated the triangulation function
supplied by the calibration toolbox.
In addition to length measurements, the three-di-
mensional coordinates extracted from the still-frame
images provided data on the position and orientation
of walleye Pollock in relat ion to the trawl (Fig. 5).
These data were used to determine distances of pollock
targets to trawl components for position of fish and to
calculate tilt and yaw for orientation of fish.
Dat a collections
Field testing of the video -drop system was conducted
12 –15 Ju ly 2008 at Zhemchug Rid ges, locat ed on
the east er n Ber ing S ea shelf adjacent to Zhemchug
Canyon where a sizable rockfish p opulation is pres-
ent in u ntrawlable and isolated rocky ridge area (Fig.
6; Rooper et al., in press). The camera s ystem was
deployed of f the side of the vessel FV Vesteraalen by a
winch suspended from a block attached to the vessel’s
crane. The camera sled dri fted with t he prevai ling
cu rrent, whi le the camera wi nc h operator kept the
seaf loor in view and avoided a ny obstacles using rea l-
time navigation . Stereo video was collected over 11
transects, each rang ing in length from 3.5 to 49.5 min
and covering d istances of 95 m to 1673 m. Obser va-
tions of trawl movements with the still-frame system
were made dur ing acoustic surveys of pollock in the
eastern Bering Sea in June and July 20 07 onboard the
RV Oscar Dyson (F ig. 6).
Testing of the c alibr ations for the t wo camera syste ms
To test the video calibration five random still images was
selected from the video-drop system of the checkerboard
taken at the begin ning and end of the study. Three
intervals of 10 cm, 20 cm and 30 cm each were measured
three times from the top to the bottom of the checker-
board (n= 3 for each interval) and averaged within each
frame. The average from each frame multiplied by the
interval combination was then tested in an analysis of
variance to determine whether there were significant
differences between measurements from the first and
second measurement set.
357
Williams et al.: Use of ster eo-camera systems i n assessing roc kfish abundance and pollock behavior
Result s
Calibration
The estimates of distance of the fish to the trawl deter-
mined with the second calibration of the drop-video
system were signi f icantly larger and more var iable
than those from the first measurement set across all
three intervals (Fig. 7). Differences between the mean
measurement s a nd known values in the s econd set
ranged from 6.6% to 8.2%. However, the 95% confidence
intervals for both sets included the actual va lues for
the intervals in all cases, and the coefficients of varia-
tion for the measurements ranged to 5.5% of the mean
value, indicating that the length measurements were
reasonably precise. A similar procedure was also per-
formed with the still-frame system, but only a single
set of validation measurements was made before the
start of f ield operations. The resu lts of this set closely
matched that of the f irst set made with the video cam-
eras (Fig. 7).
Fis h lengths determin ed with the vide o-drop system
The adult rockfish observed in the video were northern
rockfish (96.94%), unidentified adult rockfish (Sebastes
spp., 0.98 %), adult Pacif ic ocean perch (0.49 %), and
dusky rockfish (S. ciliatus,1.60%), whereas most of the
juveniles that were identified to species were Pacific
ocean perch (Rooper et al. in press). Some of the juve-
nile rockfishes observed in the v ideo were too small to
identify to species. I ndividuals of each species group
were randomly chosen to be measured in proportion with
their abundance. Up to 200 randomly selected individual
rockfish were measured in each transect, resulting in
a total of 1489 length measu rements. Rockfish were
measured by using fork length only if both the tip of
their snout and the end of the tail were plainly visible
in both still images. If the randomly chosen rockfish
could not be measured, the next available rockfish of the
same species group that was deemed measureable was
chosen. In a few cases, where the occurrence of a spe-
cies group was very small (<5 individuals in a transect),
none were measured.
A random sample of 20 rockfish that were observed in
successive still frames of both video cameras was used
to determine measurement precision and to estimate
distance of the fish from the camera. These fish were
measured in up to four consecutive frames and their
estimated length were compared by using linear regres-
sion (Fig. 8). The percent difference between successive
length measurements was not significantly related to
the average fish length (P= 0.28); in other words, there
was no length-related bias in the measurements. The
length data were also tested for a relationship with dis-
tance from the camera by using linear regression. There
was no bias in the measurements of fish for distance
from the camera (P=0.29). T he standard deviation of
fish length = distance ([Xh, Yh, Zh], [Xt, Yt, Zt])
[Xt, Yt, Zt]
[Xh, Yh, Zh]
LYt
LYh
LXh
LXt
RYt
RYh
RXh
RXt
Left
camera Right
camera
Y
Z
XFigure 3
Method for determining f ish length measurements by using s tereo images. T he th ree-dimen-
sion al coordi nat es of the fish head and tai l (Xh,Yh, Zh; Xt,Yt,Zt) were deter mi ned by stereo-
trian gulation and by usi ng the im age-based coord inates from the image pairs (i.e., L Xh,LYh;
RXh,RY h). Fish length was estimated as the Euclidia n distance between the three-dimensional
points of the he ad and ta il.
358 Fishery Bulletin 108(3)
the percent difference in multiple measurements of the
same fish was 0.076.
Analysis of still -frame ste re o i mages
Ten deployments were made with the still-frame system,
and ~200 fish were measured per deployment. Catches
consisted almost exclusively of walleye pollock (>99%). A
comparison between the length frequencies derived from
the stereo analysis (n= 360) and physical measurements
of fish captured in the codend (n=1260, Fig. 9) showed
that optical sampling approximates the length-frequency
distribution of fish caught, despite the smaller sample
size for optical sampling.
In addition to length measurements, the stereo analy-
sis prov ided data on walleye pollock orientation and
their relative position within the trawl. Quantitative
descriptions of the distribution of tilt and yaw angles
were easily calculated by using the same points in im-
ages (head and tail) derived for fish lengths (Fig. 10).
To calculate the position of fish within the trawl addi-
tional corresponding points along the trawl panel were
identified and their three-dimensional coordinates were
determined by the triangulation process outlined above
(Fig. 5).
Discussion
The potential of stereo cameras for measuring marine
organisms has been shown in many studies (i.e., Shortis
et al., 20 00 ; Harvey et al., 2003), but here we present
a description of the complete implementation of stereo
cameras, including equipment costs (Table 1), image
analysis process, and expected precision in data from
these systems. The two stereo-camera systems described
here were studied for their potential to provide infor-
mat ion to augment fisheries assessment sur veys in
Alaska. Specifically, the stereo-camera systems in our
study provided species and length data for untrawlable
regions located within bottom-trawl survey boundaries
and provide a new method for studying the behavior
of f i sh in a midwater trawl. Our main goal was to
pre sent field-tested method s to provide qua ntifiable
image-based data for fisher ies surveys and our results
may help similar research with stereo- camera–based
sampling systems.
The video - drop system was usefu l for esti mating
rockfish size and species composition in field tests in
Alaska. Error rates for size were on the order of 8.2%
or less, which equates to about 2.5 cm for a 30 -cm fish.
Compared with other studies with error rates of ~0.1%
to 0.7% in stereo-video systems (Harvey et al., 2002 ;
Harvey et al., 2003; Shortis et al., 2009), the measure-
ment error rate in our study was high. This rate repre-
sents systematic error most likely caused by the need
to remove cameras from the housing after each deploy-
ment because a sl ight misal ig nment of the cameras
in relation to the position at calibration would reduce
the precision of the measurements. Ruff et al. (1995)
report an ach ievable level of precision in measur ing
fish of 3.5%, based on repeat measurements of indi-
viduals, which is also better than the 5.9% observed
in our study. The error rates also compare well to the
rates of 1–5% for measuring rig id items with parallel
lasers (Rochet et a l., 2006). However, only fish on or
near which the parallel laser beams are projected can
be measured. This restriction limits the measurement
sample size. In contrast, any fish simultaneously viewed
by both cameras in a stereo-camera system can be mea-
sis7left_9.bmp
sis7right_9.bmp Load images
Recover data
Load Calibration File
New Target
Delete Length
Delete Mark
Measure
Mark Point
Triangulate
Load Datafile
Save Results
3D plot
Figure 4
Computerized display of the stereo processing options for the drop st ereo-video with a custom-built appl ication wr itten
in the Matlab computing lang uage. Synchronous i mages were ext racted from videos ta ken by two cameras and used to
estimate f ish leng th. Images were taken in Zhemchug Rid ges in the easter n Bering Sea in July 2008.
Left Shot Info
Right Shot Info
PAN
ZOOM
359
Williams et al.: Use of ster eo-camera systems i n assessing roc kfish abundance and pollock behavior
Figure 6
Map of study ar eas in the easter n Beri ng Sea show in g the location of f ield
tests of t he drop st ere o v ideo cameras for sampling untrawlable areas (black
square) i n July 2008 and for sampli ng fish behavior in a trawl (c ircles) with
a sti ll-frame stereo camera in July 2007.
A raw image B processed image C 3D reconstruction
height (m)
across trawl axis (m)
along trawl axis (m)
3
2
1
0
–1
5
0–2
02
Figure 5
Images of wa lleye pollock (Therag ra chalcogramma) from a st il l-f ra me ster eo camera in a midwater trawl and a 3-D
rec onstruc tion of fish in relation to the trawl net. (A) Fish leng ths were meas ur ed by enla rg ing the ima ge of a f ish
and indicatin g the position of the snout and tail (shown as U) in both right a nd left raw ima ges (on ly the left image is
show n above) . (B) The chosen f ish endpoints are overla id on the image as lines. In addition to estimates of f ish length,
stereo -pr ocessing a llows the position of f ish i n relation t o the trawl to be esti mat ed. A dditional point s in the images ca n
be determined by findi ng cor responding left-right image pixel coord inates (B, shown as crosses) . (C) Following stereo-
trian gulation, a three-dimensional plot shows the f ish ta rgets as arrows and trawl mesh k nots as dots.
sured, and thus the number of
fish that can be measured is
larger from the same leng th
transect. Improvements in the
qua lity of the still-frame im-
ages and in the collect ion of
calibration data from a target
at the beginning of each tran-
sect may allow more precise
me a s u r eme nts to be taken
in future studies. Given our
ina bility w ith ot her su r v ey
gear s t o determ ine fish siz e
and species composition in un-
trawlable habitats, the use of
stereo cameras holds promise
for stock assessments of rock-
fish and other species. Stereo-
camera-based sampling could
als o be us ed broa dly wh er-
ever gears other than bottom
trawls are needed to obta i n
speci es- or size - composit ion
information.
Lengths of rock fish derived
fr om the video-drop system
were generally comparable to
trawl catch-based size distri-
butions for the species exam-
St. Matthew Island
Zhemchug
Canyon
Zhemchug Ridges
St. Paul Island
360 Fishery Bulletin 108(3)
0
0.05
0.1
0.15
0.2
0 10 20 30 40 50 60
0
0.05
0.1
0.15
0.2
Proportion
Length (cm)
A
B
Figure 9
Comp ar ison between leng th freq uencies of wa lleye pollo ck (T her agra chal-
cog ramm a) estimated from (A) the im ages from a stil l-f rame stereo ca mera
(n=360) w ith in a midwater trawl, and leng th frequencies obtained from (B)
fi sh captured in the codend and dir ec tly measu re d (n=12 60). The smaller
ca mera-based sample results are sim ilar to the direc t measur ement s in
their overa ll size d istribution, but there was less defi nition for larger fish
(>20 cm). Data were pooled from three trawl samples taken in the east ern
Ber ing Sea in July 20 07.
−20 −16 −12 −8 −4 0 4 8 12 16 20
0
2
4
6
8
10
12
Frequency
Percent difference from mean
Figure 8
Frequenc y of perc ent differences from multiple leng th
measu rements o f n=2 0 ro ck f is h and the mean len gth
of the fish. Individual f ish were measur ed multiple
ti mes fro m a series of images ext racted fro m video
ta ken by a d rop stereo-video ca me ra . Measurements
fr om eac h i mag e were then com pa r ed to the mean
measu rement for that indi vidual t o estim at e poten-
tial me asurem ent err or. Rockf ish ob serv ation s were
coll ect ed at Zh em chug Rid ges , ea st ern Ber i ng Sea
in July 200 8.
Percent deviation
Distance measured (m)
video – 1st measurement set
video – 2nd measurement set
still-image calibration
Figure 7
Results of measur in g known d ista nces on the check-
erb oa rd (se e F ig. 2) duri ng the first and second cali-
brations of t he drop stereo video- ca mera syst em and a
still-f ra me stereo- ca mera syst em used for estimati ng
fish lengt h and study ing b eh av ior. Cal ibra ti on is a
nec es sa ry step in the use of st ereo ca meras to a llow
mea su rements to be made from images. Values repre -
sent the percent dev iation in measurements in relation
to known va lues, includi ng 95 % conf idenc e inter vals
based on f ive measu re ments.
ined. Nor thern roc kfis h leng ths
fr om ster eo-video images ta ken
along transects at Zhemchug Ridg-
es ranged from 9 to 41 cm (mean
length= 3 0.0 cm). In th ree bot -
tom-trawl surveys near the Aleu-
tian Islands, Clausen and Heifetz
(200 2) found that the mean size
of northern rockfish was 29.9 cm
and ranged from 15 to 38 cm. Ju-
venile Pacific ocean perch lengths
(2.6 cm to 2 5.0 cm) were similar
but ranged to smaller sizes than
those found for the A leut ian Is-
lands (8.3cm to 24.9 cm; Boldt and
Rooper, 2009). Lengths of juvenile
Pacific ocean perch obtained from
stereo video were also similar to
those in three experimental tows
in the Zhemchug Ridges area in
2004 and 2007 (juveniles ranged
from 10 c m to 2 5 cm) . However,
these l eng t h s w e r e me a s u r e d
from fish captured during bottom
trawl hauls, where the incidenc e
of smaller fish may have been due
to reduced catchability of smaller
indiv idual s. Although these ob-
servations are not meant to serve
361
Williams et al.: Use of ster eo-camera systems i n assessing roc kfish abundance and pollock behavior
Tilt angle (degrees) Yaw (degrees)
A B
Frequency
Figure 10
Orientiation of walleye pollock (Theragra chalcogr amma) within a midwater
trawl. A still-frame stereo camera was used to determine (A) the distribution
of ind ividua l fish tilt a ngle (dev iat ion of snout-tail axis from the horizontal),
and (B) the yaw (angle of fish heading in the horizontal plane). Most f ish
were oriented hor izonta lly and facing towa rd t he trawl op ening. Image data
were collecte d in the eastern Bering Sea in July 2007 as part of a study of
fish behav ior w ith in the trawl.
as a q uantitative comparison of
trawl- and stereo-camera–derived
size estimat es, they demonstrate
th e simila r itie s in in form ation
supplied by the two methods and
the potential for stereo cameras to
overcome some problems with the
catchability of juvenile fish.
The stereo camera was very use-
ful for studying behavior of pollock
in the trawl. The data show the
possibility of performing a length-
based analysis of behavior which
will directly contribute to studies
of gear selectivity and future de-
sig ns of scientif ic trawl gear. A l-
though a p ostsu rvey calibrat ion
was not performed with the still-
frame system, the camera s were
securely fastened in the housings
and were not removed during the
entire data collection, thus main-
taining intercamera spacing and
an gles. The agreement between
the cat ch-based leng th measure-
me n t s a nd t he s t e r e o - de r iv e d
leng ths provides direct validation
of the stereo-derived measurements. The low sampling
frequency of 1 frame per 5 s ensured minimal inf lu-
enc e of t he arti fic ial lightin g from the cameras on
behavior because the fish photographed had not been
previously exposed to the light source.
Recent development of hig h-resolution dig ital im-
aging devices and an increased access to custom de-
signed, freely available software tools have made ste -
reo- camera methods easy to implement by rese arch
gr oups w it hout direct expertise in the subject. The
camera calibration toolbox (Boug uet, 200 8) provided
the basis for software development. Although the cur-
rent analysis approach is stil l fairly t ime intensive,
the volumes of data analyzed were not very large. In a
routine application of stereo-v ideo cameras in untraw-
lable areas, additional levels of automated processing
would likely be required because the quantity of v ideo
footage wou ld subs tantial. For some aspe cts of the
analysis, such as the matching of targets on the stereo
cameras and the extraction of fish leng ths, automation
may be atta inable, whereas automating more dif ficult
tasks of isolating a nd identifying fish targets may not
be feasible.
Stereo photography will continue to be developed as
sur vey tools are developed for monitoring fish stocks
and thereby improving the quality of stock assessments
of fishery resources in Alaska. Some challenges remain;
for instance, the challenge of institutionalizing image-
based sampling as a routine survey method for untraw-
lable habitats. As a method of studying fish behavior
in trawls, stereo cameras provide promising results by
allowing three-dimensional reconstructions of the trawl
environment.
Acknowledgments
The development and deployment of the c amera and
winch system was possible only through the assistance
of G. McMurr in, G. Mundell, B. Lauth, G. Hoff, and
especially S. McEntire. T. Cosgrove, K. Sjong, L. Mavar,
and M. Booth of the FV Vesteraalen were instrumental
in conducting the field t ests. Calibration and deploy-
ment of the still-frame camera system were significantly
aided by D. Jones and A. McCarthy. This manuscript
was reviewed by D. Somerton, G. Hoff, R. Lauth, and
M. Wilkins.
Literature cited
Boldt, J. L., and C. N. Rooper.
200 9. A bund a nce , co ndit io n , and die t o f ju ven i le
Pacific ocea n perch (Seb ast es alutu s) in the A leutia n
Islands. Fish. Bull.:107: 278 –285.
Bouguet, J. Y.
2008. Camera cal ibration toolbox for Matlab [onl ine].
[Available from http://vision.caltech.edu /bouguetj/calib_
doc /index.html (acc essed S ept ember 2008)] .
Clausen, D. M., and J. Heifetz.
20 02 . The northern rockf ish, Seba st es polys pini s, in
Alask a: commercia l f ishery, distr ibution, and biolog y.
Mar. F ish. Rev. 6 4:1–2 8.
Cordue, P.
20 07. A note on non-r andom error str uc tu re i n trawl
survey abundance indices. ICE S J. Mar. Sci. 64:1333 –
1337.
Godo, O. R., W. A. Karp, and A. Totland.
199 8. Effects of t rawl sa mpling variability on precision
of acou stic abunda nce estimat es of gadoids from the
362 Fishery Bulletin 108(3)
Barents S ea and the Gulf of Alaska. ICES J. Mar.
Sci. 55 :86 –94.
Gunderson, D. R., and T. M. Sample.
1980. D istribution a nd abundance of rockfi sh off Wash-
ington, Oregon and California du ring 1977. Mar. Fish.
Rev. 42: 2–16.
Harvey, E. S., M. Cappo, M. R. Shortis, S. Robson, J. Buchanan,
and P. Speare.
2003. The accuracy and precision of underwater measure-
ments of leng th a nd ma xi mum body depth of southern
bluefi n tuna (Thunnus maccoy ii) w ith a stereo-video
camera syst em. Fi sh. Res. 63 :315–326.
Harvey, E. S., D. Fletcher, and M. R. Shortis.
2002. Esti mat ion of reef fish lengt h by diver s and by
stere o-video : a f irst compa rison of the accuracy and
precision in the f ield on living fish under operational
conditions. F ish. Re s. 57: 255–265.
Karp, W. A., and G. E. Walters.
199 4. Survey assessment of sem i-p ela gic Gadoids: the
example of walleye pollock, Theragra chalcog ramma, in
the East ern Ber ing Sea. M ar. Fish . Rev. 56: 8 –22.
Rochet, M. J., J. F. Cadiou, and V. M. Trenkel.
2006. Precision and accuracy of fish length measurements
obtained w ith two v isual underwater methods. F ish.
Bul l. 104:1–9.
Rooper, C. N., J. L. Boldt, and M. Zimmermann.
20 07. A n assessment of juve ni le Paci f ic ocean perch
(Sebastes alutus) habitat use in a deepwater nursery. Es-
tua r. Coa st. Shel f Sc i. 75 :371– 380.
Rooper, C. N., G. H. Hoff, and A. DeRobertis.
In press. A ssessing habitat ut ilization and rock fish bio-
mass on an isolated rocky ridge using acoust ics and
stereo i mage analysis. Can. J. Fish. Aquat. Sci.
Ruff, B. P., J. A. Marchant, and A. R. Frost.
1995. Fish sizing and monitoring usi ng ster eo image
ana lysis system applied to fi sh farming. Aquat. Engi-
neer. 14 :155–173.
Shortis, M. R ., S. Miller, E. S. Harvey, S. Robson.
2000. An ana lysis of the calibration stability a nd mea-
su rement a ccu rac y of an u nderwate r st er e o-v id eo
system used for shellf ish sur veys. G eomatics Res .
Aust. 73 :1–24.
Shortis, M. R., J. W.Seager, A. Williams, B. A. Barker, and
M. Sherlock.
2009. Using ster eo video for deep water benth ic habitat
sur veys. Mar. Tech. Soc. J. 42:2 8–37.
Van Rooij, J. M, and J. J. Videler.
1996 . A simple field met ho d for stere o-phot og raph ic
leng th measurement of free swi mming fish: merits and
constraints. J. Exp. Mar. Biol. Ec ol. 195:237–249.
Xu G., and Z. Zhang.
199 6. Epipola r geometry in st ere o, motion, and objec t
rec ognit ion: a u ni fied approach, 336 p. Kluwer Aca-
demic Publs., Norwell, MA.
Zimmermann, M.
2003. Calculation of untrawlable areas within the bound-
aries of a bottom trawl survey. Ca n. J. Fish. Aquat.
Sci. 60 : 657–669.
