For many of us, Where’s Waldo? was a childhood staple. Hours were spent poring over pages of Martin Handford’s books, our developing eyes and brains desperately searching for the image of a brown-haired, glasses wearing gent wearing a white and red cap. Now, there’s a robot that can do this for you. It’s called “There’s Waldo.”
“There’s Waldo” was built by Redpepper, a creative ad agency. The robot uses facial recognition to spot Waldo’s visage in classic “Where’s Waldo?” images, spotting the sneaky glasses-wearing fellow in as fast as 4.5 seconds. Once identified, a weird motorized rubber hand slaps itself down atop where Waldo is located on the page.
There’s Waldo snaps a photo of the page, uses OpenCV to find and extract the faces in the photo, and then employs Google’s Auto ML Vision service to compare each face to the image of Waldo the machine has been trained on. When it’s at least 95% confident it’s made a match, the robot moves its arm to point at Waldo’s coordinates on the page. The robotic arm (a uArm Swift Pro) is controlled by a Raspberry Pi processor.
In order to successfully train the machine’s algorithm, its creator used 62 different Waldo heads and 45 Waldo head and body images. Matt Reed, Redpepper’s Creative Technologist, told the Verge that he built There’s Waldo with no previous AutoML experience. It took him about a week to program using Python.
While There’s Waldo doesn’t really serve any practical purpose—as far as we’re aware, there’s no professional Waldo-finding league out there—it is a fun proof of concept. We wouldn’t want to put it directly in a contest with any young kids, though. There’s Waldo is so speedy the competition would likely just end with tears or a tantrum, from its frustrated human competitors.