Biol. Bull. 211: 101-105. (October 2006)
© 2006 Marine Biological Laboratory
Adaptation of Underwater Video for Near-Substratum Current Measurement
Russell C. Wyeth* and
A. O. Dennis Willows
Department of Biology, University of Washington, Seattle, Washington 98195-1800; and Friday Harbor Laboratories, 620 University Road, Friday Harbor, Washington 98250
* To whom correspondence should be addressed, at Department of Physiology and Biophysics, Dalhousie University, Halifax, NS, B3H 1X5, Canada. Russell.Wyeth{at}dal.ca
Abbreviations: PTV, particle tracking velocimetry • RTF, relative to flow
We describe a new method to measure current near the substratum. A variety of current meters are available to record flow data relevant to benthic organisms in their natural habitat. Several factors, including accuracy and precision, sample volume, sample rate, deployment constraints, and cost make them more or less appropriate to different applications. Our method, based on tracking particles in videos from a camera mounted near the substratum, provides an inexpensive option for measuring flow. We validate this "current camera" against a conventional electromagnetic meter, and then use it successfully to estimate flows experienced by the nudibranch Tritonia diomedea in its natural habitat. As we implement it, this current meter samples a large volume and provides a measure of bulk flow in the habitat. However, the method is adaptable to videos acquired from other cameras at a variety of size scales and locations. We therefore conclude that our method is an option that should be considered by biologists interested in measuring flow.
In a recent field study (1), we used current direction to partially explain navigation behaviors of the nudibranch mollusc Tritonia diomedea Bergh. To avoid interference with the behaviors and our ability to observe them, we chose to measure flow some distance away from the slugs. As a consequence of this constraint, we needed integrated current headings (over time or space) to overcome the heterogeneous nature of flow in this habitat and thus allow us to estimate flow in nearby regions. Furthermore, we needed measurements made near the substratum, to correspond with flow experienced by the benthic slugs. Several current meter options are available. These include acoustic doppler velocimeters (ADVs), acoustic doppler current profilers (ADCPs), and electromagnetic meters. However, none of these are ideal. ADVs have a very small sample volume, near 1 ml (2). ADCPs, although adapted for near-substratum use (3), still have relatively small sample volumes, typically less than 1 liter (4). Electromagnetic meters have either small sample volumes or are compromised when deployed close to the substratum (pers. comm: Marsh-McBirney, Inc., Frederick, MD; Interocean systems, Inc., San Diego, CA). All are expensive. Two other cruder options, mechanical meters and dye tracking, are less accurate. We therefore chose to use particle tracking velocimetry (PTV) in videos from a camera placed to record particle movement near the substratum (Fig. 1). This "current camera" has several advantages. (1) PTV is an established method for determining fluid flow directions (5–7). (2) A large sample volume (Fig. 1) can be created with the appropriate camera field of view. (3) Current meter output is not corrupted by proximity to the substratum. (4) The meter is unobtrusive (camera dimensions: 5 x 5 x 8 cm; pole diameter: 2.2 cm) and will have little effect on flow. (5) The system requires the same hardware for deployment and data capture as typically used to record slug behaviors and thus is easily integrated into field studies. (6) All current meters are susceptible to fouling and other disruptions, and a video record gives unequivocal information on the source of anomalous data.
View larger version (22K):
[in this window]
[in a new window]
|
Figure 1. Method used to determine flow direction from videos of particle movement. (A) Schematic of the "current camera," not to scale. A video camera (c) was attached to a pole and pointed down toward a black metal cookie sheet (ms; 38 x 25 cm) placed flat on the substratum (su) below. This arrangement recorded particles moving in a volume of water (estimate: 141, outlined by dashed lines) between the camera and the sheet. Video, digitized at 30 frames s–1 (fps), was subsequently processed in 5-s clips (comprising 150 sequential frame pairs). (B) A sequential pair (i and ii) of frames from a current video shows white particles against the black background, some of which are suspended in the flow (one circled). The same particle is circled in panels C–D showing the results of video processing. (C) Frames were cropped to the metal sheet boundaries, and moving-particle contrast was enhanced by subtracting the mean projection image for the entire clip from each frame (11). (D) Each enhanced frame was converted to a binary image with between 5 and 20 white particles by using an adaptive gray-level threshold. Contrast fluctuation artifacts could be distinguished from real particles based on both size and shape and were excluded from processing. (E) A particle tracking velocimetry algorithm (PTV, Fig. 2) then attempted to link the same particle between one frame and the next, while excluding links to other particles. This composite image shows particles from this frame pair (1st frame, white; 2nd frame, black) with link vectors contributing to the heading calculation (white lines; mean shown by central arrow) and vectors of comparable length but excluded from the calculation (black lines). Video from the camera (model CVC-320WP, Speco Technologies, Amityville, NY) was recorded by a digital video system (Novex 2000, Novex (Canada) Ltd., North York, ON, Canada) and saved as AVI files with the Indeo5 compression algorithm. Processing was fully automated in Matlab (The Mathworks Inc., Natick, MA). Scale bars, 10 cm.
|
|
The current camera recorded particles moving in flow across a black background (Fig. 1). To convert particle motion into flow measurements, we designed an automated particle tracking velocimetry algorithm (Fig. 2; Movie 1 [http://www.biolbull.org/supplemental/]; code is available in the online supplementary material [http://www.biolbull.org/supplemental/or by contacting the corresponding author]). In short, the algorithm identified moving particles in successive video frames, and then searched the pool of vectors linking particles between frames for a set with consistent length and heading. The algorithm was designed to accommodate various particle speeds, since the videos recorded particles moving at different levels within the substratum boundary layer (8). For this same reason, we did not calculate flow speed. The algorithm calculated a mean heading from every 5 s of video, averaging over space (between each pair of frames, headings for 5 to 20 particles were averaged) and time (mean headings from 150 frame pairs were averaged). Qualitatively, the calculated flow direction corresponded with the movement of particles in the current videos (Movie 1 [http://www.biolbull.org/supplemental/]). This was true even when flow changed during the 5-s video clip. For example, in oscillatory flow due to wave action, the algorithm tracked the oscillation, and thus calculated an accurate mean heading for the entire clip (Movie 2 [http://www.biolbull.org/supplemental/]). In addition, the algorithm prompted for human intervention when the uncertainty associated with the final mean heading exceeded a conservative threshold (Fig. 2). A human observer then corroborated or overrode the algorithm output, as necessary. Thus, flow measurement continued in the rare cases when fish or drift algae interfered with the automatic operation of the current meter.
View larger version (30K):
[in this window]
[in a new window]
|
Figure 2. Particle tracking velocimetry algorithm used to determine flow headings. A mean heading was determined for every 5 s of video by tracking particles between 150 successive frame pairs. The six steps in the algorithm attempted to link the same particles between one frame and the next, while excluding links to other particles. The algorithm was automated in Matlab (The Mathworks Inc., Natick, MA) and is available, along with parameter details, in the online supplementary material [http://www.biolbull.org/supplemental/]. Any clip without a strong final mean heading (Hotelling test, P > 10–20; 12) was checked by a human observer who could adjust the flow measurement if necessary.
|
|
We used several methods to validate the current camera’s performance. We found that the algorithm output compared favorably to mean flow headings calculated by tracking particles manually in the video (unpubl. data), in blind comparisons with human observers of the video (unpubl. data), and to measurements made by a conventional electromagnetic current meter adjacent to the current camera (Fig. 3).
View larger version (31K):
[in this window]
[in a new window]
|
Figure 3. Comparison of current camera with a conventional current meter. We deployed the current camera next to a conventional electromagnetic current meter underneath the dock at Friday Harbor Laboratories. (A) An hour of electromagnetic meter output (solid line) and current camera output (X) shows the strong similarity between the two current meters (B and C) For all 37 h of simultaneous measurements, we used a visual procedure to demonstrate agreement between the two instruments (13,14), since there are problems with testing for similarity (15), especially with circular data. (B) We plotted the mean of the two instruments’ measurements (the mean estimates the true flow direction) against the difference between the measurements to show that (1) there was no average difference in flow direction measured by the camera relative to the electromagnetic meter (mean difference = 0.43°, dashed line), and (2) the pattern of differences across all flow directions showed little direction-specific bias between the instruments. (C) Furthermore, the histogram of differences is densely clustered and symmetrical around unity when pooled across all directions, with 95% of the differences falling within ± 61° (dotted lines in B). We placed the electromagnetic meter in 4 locations around the current camera to mitigate differences that arise from the only partially overlapping and differently sized sample volumes (current camera = 14 l; electromagnetic <0.5 1). The data for (B) were also restricted to points with electromagnetic speed output >2 cm s–1 since the video showed the current meandered extensively at lower speeds and we would not expect the two meters to have the same output. Given these constraints and the complicated flow environment underneath the dock created by pilings and debris, we suggest that the two instruments show strong similarity, and thus the camera and particle tracking velocimetry algorithm form a valid method to measure flow. Electromagnetic meter: Marsh McBirney Inc., model 511, Frederick, MD, with output digitized at 60 Hz using a Micro1401plus ADC with Spike2 4.x software, Cambridge Electronic Design Ltd., Cambridge, England. Headings averaged every 5 s to correspond to current camera output.
|
|
For our study comparing flow direction and navigation in T. diomedea (1), we used three experiments to show that current camera measurements accurately estimated flow in the nearby region where behaviors were recorded. First, we found the output of two current cameras (separated by a distance greater than that between any slug and a current camera) to have short-term differences but no long-term average bias (9). Second, fluorescein plumes, with few exceptions, followed the current camera algorithm output (Movie 3 [http://www.biolbull.org/supplemental/]). Third, when slugs were swept off the substratum by flow, their subsequent direction of movement was measured as downstream by the current camera data (Fig. 4; Rayleigh test, mean heading 
= 187° relative to flow, r = 0.93, z18 = 15.5, P = 0.000012). We therefore conclude that current camera videos analyzed by our algorithm are an effective current meter for measuring flow experienced by T. diomedea individuals.
View larger version (10K):
[in this window]
[in a new window]
|
Figure 4. Passive downstream movement of Tritonia diomedea confirms accuracy of current camera measurements. We measured the direction of movement of T. diomedea individuals drifting downstream (after dislodgement or swimming) in videos recorded 1–4 m away from the current camera. By calibrating both slug and current videos with a compass, we were able to test whether the passive downstream motion of the slugs was measured as downstream by the current camera. Each point indicates a mean slug heading relative to flow (RTF) measured by the current camera (top arrowhead, upstream (000° RTF); central arrow, mean heading RTF (, r) for all points; grey sector, 95% confidence limits on true mean heading RTF; circle, r = 1). We conclude that the current cameras indicate flow experienced by T. diomedea, because the 95% confidence limits for the mean heading RTF in the current cameras include the known downstream (180° RTF) movement of these slugs.
|
|
This current meter can also be adapted for other applications. The duration and frequency of measurements are set by the capacity and frame rate of the video recording system. We were able to record many hours of measurements at 5-s intervals with our system, but there are a variety of other options commercially available. Single-plane illumination of the moving particles (perhaps achieved through some creativity with a glass rod and a laser pointer) would allow current speed calculations and analysis using particle image velocimetry (10). In addition, different cameras could capture particle motion in any number of situations, all of which could be tracked by the PTV algorithm presented here. Furthermore, the motion-contrast enhancement of videos prior to PTV allows greater flexibility in illumination than can be afforded if raw particle movement videos are analyzed. Finally, the PTV algorithm can be easily modified to suit different video inputs and different measurement outputs. In conclusion, we suggest that flexibility and low cost (requiring only a camera, a video recording system, and a Matlab site license) make this current meter a useful option for biologists, especially those requiring simultaneous flow and video data.
 |
Acknowledgments
|
|---|
We are grateful to D. Young and C. Staude for exemplary computer support and to two anonymous reviewers for highly constructive comments. We also thank O. M. Woodward, G. Yahel, M. Baltzley, A. Vidal, S. D. Cain, W. Moody, R. R. Strathmann, G. VanBlaricom, T. Daniel, and the staff of Friday Harbor Laboratories (FHL) for help or suggestions that contributed to this manuscript. Support was provided by the Packard Foundation. R. C. W. thanks C. P. Holmes, "harem j" of FHL, and acknowledges support from the National Sciences and Engineering Research Council (Canada) and the Conchologists of America.
|
Footnotes
|
|---|
Received 8 June 2005; accepted 11 July 2006.
|
Literature Cited
|
|---|
 TOP
 Literature Cited
|
|---|
- Wyeth, R. C., O. M. Woodward, and A. O. D. Willows. 2006. Orientation and navigation relative to water flow, prey, conspecifics and predators by the nudibranch mollusc Tritonia diomedea. Biol. Bull. 210:97–108.
- Nortek AS. 2006. Vector velocimeter high resolution 3D current meter. Product brochure. [Online]. Available: http://www.nortek-as.com/brochures/Vector.pdf [accessed 2006, 7 March].
- Mudge, T. 2004. Introduction and reference guide for the SonTek/YSIpPulse-coherent acoustic doppler profiler. Product brochure. [Online]. Available: http://www.sontek.com/papers/pcadpref/pcadpref.htm [accessed 2006, 7 March].
- Nortek AS. 2006. Aquadopp® profiler. A small and light current profiler for coastal measurements. Product brochure. [Online]. Available: http://www.nortek-as.com/brochures/AquadoppProfiler.pdf [accessed 2006, 7 March].
- Adamczyk, A. A., and L. Rimai. 1988. Two-dimensional particle tracking velocimetry (PTV): technique and image processing algorithms. Exp. Fluids 6:373–380.
- Khalighi, B., and Y. H. Lee. 1989. Particle tracking velocimetry: an automatic image-processing algorithm. Appl. Optics 28:4328–4332.
- Breithaupt, T., and J. Ayers. 1998. Visualization and quantification of biological flow fields through video-based digital motion-analysis techniques. Mar. Freshw. Behav. Physiol. 31:55–61.
- Nowell, A. R. M., and P. A. Jumars. 1984. Flow environments of aquatic benthos. Annu. Rev. Ecol. Syst. 15:303–328.[ISI]
- Wyeth, R. C. 2004. Odors and water flow guide navigation in Tritonia diomedea. Ph.D. dissertation, University of Washington, Seattle, WA.
- Prasad, A. K. 2000. Particle image velocimetry. Curr. Sci. 79:51–60.
- Wyeth, R. C., and A. O. D. Willows. 2006. Field behavior of the nudibranch mollusc Tritonia diomedea. Biol. Bull. 210:81–96.
- Zar, J. H. 1999. Biostatistical Analysis, 4th ed. Prentice-Hall, Upper Saddle River, NJ.
- Bland, J. M., and D. G. Altman. 1999. Measuring agreement in method comparison studies. Stat. Methods Med. Res. 8:135–160.[Abstract/Free Full Text]
- Bland, J. M., and D. G. Altman. 1986. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1:307–310.[ISI][Medline]
- Parkhurst, D. F. 2001. Statistical significance tests: equivalence and reverse tests should reduce misinterpretation. Bioscience 51:1051–1057.
