• Smartphones make identifying endangered animals easy

    Everyday mobile tech takes the legwork out of tracking hard-to-find animals, and makes life easier for field biologists

    Smile, you're on camera <i>(Image: Ralph Clevenger/Corbis)</i>

    Smile, you're on camera (Image: Ralph Clevenger/Corbis)

    FIELD biology is notoriously laborious. In one famous 1982 study, Smithsonian entomologist Terry Erwin counted by hand the number of insect species in one hectare of forest canopy in Panama. By extrapolation Erwin estimated the total biodiversity of Earth's insects. What if your smartphone could take some of the strain?

    Harvard biologist and computer scientist Walter Scheirer has devised a machine vision system that automatically recognises and counts specific animals and runs on a Motorola Droid X2 smartphone. This will help biologists make quicker, more accurate judgements about the health of delicate ecosystems.

    Two years ago, Edwards Air Force Base in the Mojave desert, California, put out a call for a cheap way to monitor the animals that live there. The area is one of the last refuges for the endangered desert tortoises as well as the threatened Mojave ground squirrels (see map). Keeping an eye on the health of the animal population in such a remote location is time-consuming and expensive. So Scheirer developed detection and classification algorithms capable of identifying tortoises and squirrels with nothing more than a standard smartphone.

    Automated camera traps already exist, but they are not selective enough. "Right now, we have to manually go through every photo to identify species and separate photos of interest from false photos. It's a very laborious task," says Princeton conservation biologist Siva Sundaresan, who works with Grévy's zebras in Kenya. He says Scheirer's method is potentially very useful to biologists.

    But how does a phone tell the difference between a squirrel and a rock or a tumbleweed? Scheirer's system starts off by scanning its environment for objects that could be the animals it wants. It looks for clumps of pixels that are new to the scene, then examines them to decide whether they represent any of the animals it has been trained to recognise. Rather than checking each individual pixel, Scheirer's algorithms analyse the content of a frame of video and look for patterns of pixels that identify the animal. The algorithms don't need intensive processing and so run well on smartphones.

    A paper due to be presented at the Workshop on the Applications of Computer Vision in Clearwater, Florida, later this month shows how well the algorithms work, with the system able to distinguish between three different species of ground squirrel 78 per cent of the time, even though they are almost identical. Scheirer says that the algorithms have been tweaked and that the recognition rate is now around 85 per cent.

    Scheirer says his goal is to build a cheap, easy-to-use system that can automatically detect animals in any environment. More field trials are scheduled for next month, and the team aims to deliver a finished system to the US Air Force by 2014.

    Princeton population biologist Dan Rubenstein says machine vision systems will also help us understand delicate ecosystems in finer detail. "You won't be generalising from such small scale to such a massive scale," he says. "We're going to be able to save ecosystems."

    Another system presented at the conference, called Hotspotter, recognises individual animals like zebras and giraffes by their stripes and spots, although it still needs some human guidance, unlike the Mojave desert system. Rubenstein, who works on Hotspotter, says that systems like this will allow biologists to look at animals and their actions on an individual basis. "We could start to build massive databases of who's who, and how they move in time. We can use social networks to figure out how they relate to each other."


  • Commentaires

    Aucun commentaire pour le moment

    Suivre le flux RSS des commentaires

    Ajouter un commentaire

    Nom / Pseudo :

    E-mail (facultatif) :

    Site Web (facultatif) :

    Commentaire :