Object-recognition software adds green boxes when any
penguin is recognized and yellow boxes when a specific penguin is recognized.
Credit: Tilo Burghardt
“You can encode the animal pattern even more effectively and efficiently than human faces,” he says. “You don’t have to use a lot of description to make the system work.”
The system uses fairly cheap components: ordinary security cameras connected to laptops, which communicate via a wireless LAN. With a power source and a connection to stream the data to a central server, it operates in the field with minimal human interference. In a month of observation, Barham says, the system will capture data on almost the entire colony.
The challenge in generalizing this approach to other species, of course, is in simply collecting the images efficiently. For wide-ranging species that don’t travel along well-used paths, passive cameras won’t capture enough images to track an entire population.
But even for mobile animals that can’t be photographed passively, object-recognition software can take the place of the painstaking work of hand-matching images, a job that takes great expertise and eats up limited conservation research budgets. Sophie Grange, a zebra biologist at Wits University, in South Africa, is optimistic about the technology’s potential, and she is currently working with Burghardt and his colleagues to develop a similar system for her fieldwork. “These studies are essential to improving our scientific knowledge on animal demography, which is central if you want to manage and preserve animal populations,” she says.
Burghardt thinks that the field of conservation biology is ripe for technological innovation. “It took a long time to realize you can use similar technology to solve seemingly very different problems,” he says. “We’ve basically opened up a new field of collaboration between science and engineering.”