The robot apocalypse isn’t upon us just yet. Even though drones are still way better than people at taking pictures of wildlife populations, they haven’t yet supplanted us in the realm of aerial image interpretation. And the reason for this carries a bit of irony; drones are so efficient at collecting massive amounts of data in the form of aerial pictures and videos, that processing it all can quickly become an overwhelming task, even for a computer.
This why researchers at Switzerland’s École Polytechnique Fédérale de Lausanne took a democratic approach to aerial image processing. The team traveled to the Kuzikus wildlife reserve in Namibia and used drones to survey the resident wildlife population, including ostriches, kudus, wildebeests, zebras and rhinos. Unsurprisingly, the team ended up with a huge amount of data and their computers had a tough time distinguishing between rocks, bushes, and actual animals.
So, they turned to online crowdsourcing platform micromappers.org.
“Within two days, they had evaluated 98% of the 26,000 images that had been uploaded,” the study’s senior author, Stéphane Joost, told phys.org.
Despite some false positives, the volunteers annotated images were used to train an algorithm to recognize animals in the remaining images.
This is not a new phenomenon – in addition to African countries, drones have been used to for years survey wildlife populations in Canada, Australia, and other countries all over the world . But data collection is not the problem. Data analysis is the piece of the puzzle that still needs perfecting.
As algorithms like this are fine-tuned, drones will be able to work seamlessly with computers to accurately process aerial data as it is gathered… and yes, eventually human eyeballs will not be necessary until an actionable report is produced.