Professor E. Harvey with stereo video rig. (Photo: Dianne McLean, uwa.edu/fisheries.nsw.gov.au)
Scientists create algorithm to count fish more effectively
Friday, January 06, 2012, 04:00 (GMT + 9)
A group of Australian researchers hopes to automatically count and measure fish in remote locations around Australia and New Zealand using computer software instead of human effort.
The team has spent thousands of hours looking at underwater video of the fish.
Associate Professor Euan Harvey, who works with the University of Western Australia’s (UWA) School of Plant Biology and the Oceans Institute, and his colleagues received a three-year, AUD 450,000 (USD 465,669) Australian Research Council Linkage grant to create a computer algorithm that will count and measure fish.
The team currently films schools of fish by placing a stereo-video system in a frame similar to a craypot and dropping it to the ocean floor with a bag of bait to attract fish.
The idea is to develop the algorithm so people will not be needed to count and measure the fish on resulting video, which is tremendously time-consuming as well as expensive.
“At the moment, depending on where we are dropping the cameras and how many species and individual fish there are, it will take between two and three and a half hours to process one video,” explained Harvey. “If you put that in dollar terms, that’s probably about an extra AUD 100 (USD 103.48) per deployment for someone to sit there and analyse them.”
Harvey and his research partners record up to 2,000 hours of fish video a year.
“At the moment, we’ve got about 28 stereo video systems, and they could be in use anywhere from the Kimberley and the North-West Shelf to Guam or Hawaii,” he says.
“What we do is go to an area and deploy up to 10 cameras at any one time for an hour, pick them up and then move them to another site, so we get broad spatial coverage.
“We’re working with the Fisheries Department, the Commonwealth Scientific and Industrial Research Organisation (CSIRO) and the Australian Institute of Marine Sciences (AIMS), the National Oceanic and Atmospheric Administration (NOAA), the National Museum in New Zealand and staff and students from other universities, and the cameras will go to wherever the work is based,” Harvey said.
Mark Shortis from RMIT University in Melbourne pioneered the use of stereo-video to monitor and measure fish stocks nearly 20 years ago.
“That was the key thing, because using basic trigonometry you can calculate x, y, and z points – if you’ve got two points you can calculate the distance between them and that means you can measure the length of fish very accurately,” Harvey told.
The new algorithm project would facilitate the efforts of marine agencies to monitor finfish communities and improve husbandry in aquaculture.
Work on developing the algorithm could begin by February or March, he said.
By Natalia Real