Create a free account to continue

Untagged: Software recognises animals it's seen before

When identifying individual animals in the wild, there's a limit to what we humans can do – but now there's a way to track them by their markings

LEOPARDS can't change their spots - and tigers, zebras and whale sharks can't change their stripes. This just as well since they can be used to identify individual animals from pictures or video instead of conventional identity bands and radio tags.

It is more effective to track animals by such "fingerprints" since they don't have to be caught and sedated, which is stressful.

To monitor the population and movement of leatherback sea turtles, marine biologists normally use plastic "cattle tags", which contain a unique identity number. These often fall off, says Scott Eckert, a director of the Caribbean sea turtle conservation network Widecast. "We get a large number of leatherbacks coming into Trinidad with tag scars. So we know they've been tagged, but we have no idea when," he says. "We lose a tremendous amount of valuable information."

Widecast approached Eric Pauwels and colleagues at the Dutch centre for mathematics and computing (CWI) in Amsterdam to develop an algorithm that would identify individual leatherback sea turtles from digital images of their pink spot, a patch without pigment on the top of their heads. These spots are unique to each turtle, and don't change throughout their lives, says Pauwels. The algorithm looks for specific identifiers on the spots - in particular areas of contrast, such as white patches on dark areas. It then encodes these details into a unique biometric sign for the turtle.

The team has tested the system on digital photographs of leatherbacks, and they now plan to add it to Obis-Seamap, a global marine animal database. Biologists will then be able to upload images of sea turtles alongside details of where they have been spotted, and get an instant ID if its details are there.

It's not just researchers who could use the system, says Pauwels. "Ideally, anyone who comes across a turtle on a beach could take a photograph, upload it to the website and find out whether the animal has been seen before, where it is from, and how old it is," he says.

Anyone who sees a turtle on a beach could take a photograph and find out where the animal is from

Tilo Burghardt at the University of Bristol, UK, is hoping to do the same for African penguins. His team has developed a system to automatically identify individual penguins from video of a colony on Robben Island, off the coast of Cape Town, South Africa.

Usually, studying penguin populations means attaching numbered bands to their flippers, says Burghardt, but these can cause severe injuries. And practically speaking there's a limit to how many uncooperative birds one researcher can band.

Burghardt maps the placement and shape of spots on each penguin's chest as they pass the camera, and compares this with a database of known birds. Soon visitors to the island will be able to upload their own pictures to help with identification.

At a computer vision conference in Istanbul, Turkey, last month, Burghardt revealed that the penguin-spotting technique can be used to fingerprint any animal with spots or stripes. Next, he hopes to use it to spot great white sharks by the jagged pattern on the rear of their first dorsal fin.

It is not just an animal's surface features that can be used as identifiers: footprints are also unique, says Zoe Jewell of Belize based conservation organisation WildTrack.

While monitoring black rhino using radio collars in Zimbabwe in the 1990s, Jewell and her husband Sky Alibhai discovered that repeatedly sedating females to attach and then maintain collars reduced their fertility rates (Journal of Zoology, vol 253, p 333).

So they began to investigate whether technology could emulate the way bush trackers identify animals by their prints. The footprint identification software they developed measures the distances and angles between various landmarks on the print, to create a unique biometric signature. So far it works for white and black rhino, cheetahs and polar bears.

The software doesn't cope too well with smudged footprints, which have to be picked out by hand, but the pair are working with software firm SAS, based in Cary, North Carolina, to solve this.

WildTrack is working with a conservation group in India to help them track Bengal tigers. And they recently received a request for help in identifying a rogue elephant in a herd in Botswana that has been trampling crops. This is likely to be tough, says Alibhai, because there are likely to be few features to look out for with an elephant print.

"You can spin the thing around and you don't know where the top and bottom are," he says. But he is still optimistic that the new technique will help spot their elephant in the crowd.

When this article was first posted, WildTrack was misspelled

Birds-eye viewer

HOW would you like to perch for hours on a cold and windy cliff top, monitoring the comings and goings of hundreds of identical birds? That's often the unenviable task of those studying nesting guillemots. But help is at hand.

Patrick Dickinson of the University of Lincoln, UK, says a researcher may wish to monitor a pair of birds every day to see if their chick is still there. "But they can't stare at one pair of birds for the whole day."

So Dickinson and colleagues are developing software that can pick out birds from a background of moving foliage. Not only will this automate the process of counting the birds, but it should also give biologists valuable insights into their behaviour, says Dickinson. "One of the things they are interested in is chick survival rates, and how that correlates with the amount of time birds spend at their nests," he says. "To get data like that manually is virtually impossible."

The system will be tested on video footage of nesting guillemots on Skomer off the coast of Wales, UK. The first step is to train the software on images of the nesting area, to develop a model of the moving background. It then compares this to each new video frame, breaking the image down into different regions and looking for new objects with the right shape and position to be nesting birds.