A Harvard scientist has developed a system that uses smartphones to help track and count endangered species. The system helps get round the fact that visual recognition usually requires immense computing power.
While a computer can generally process data more quickly than the human brain, people are much better skilled when it comes to recognition. For example, when we see a friend in the street we can usually instantly recognise them even if they have a new haircut or have picked up a scar on their face.
As New Scientist explains, a computer struggles to respond so quickly as it works by comparing individual pixels rather than being able to make a snap judgment using the “big picture.” This means that while smartphones can look for a very specific image (such as a facial recognition lock), they don’t have the capacity for extensive on-board image analysis.
Walter Scheirer of Harvard tried to tackle that problem while trying to find more effective ways to count the number of a particular species in an area under study. At the moment scientists use a string of cameras that take shots when they detect movement, but humans need to look through each image to spot whether the species under study is present.
Scheirer, who works in both biology and computer science, has developed an algorithm that takes a similar approach to humans: instead of analyzing individual pixels, it looks for familiar patterns. That reduces the hardware demands to the point that it can work on smartphones, which are much more practical for leaving unattended in the natural habitat. The current version of the system uses a Motorola Droid X2.
An initial study showed that in 78 percent of cases, the system could correctly spot the difference between three species of squirrel that are almost identical in appearance. Scheirer has since refined that to an 85 percent accuracy rate and hopes to have a fully working system ready next year.