If you've spent countless hours pouring over the pages of Martin Handford's Where's Waldo books looking for a certain bespectacled wanderer clad in red and white stripes, there's now some AI to make finding Waldo much quicker.
Creative agency Redpepper has designed a new system called There's Waldo, which is capable of finding the elusive character in mere seconds.
As reported by The Verge, There's Waldo utilizes a smorgasbord of existing technology to make it work. First, there's a uArm Swift Pro that's controlled with a Raspberry Pi. The uArm is equipped with a Vision Camera Kit that's designed for facial recognition software.
When the book is open, the camera takes a photo of the page. Then, OpenCV looks at the dozens of cartoon faces, finding all possible Waldos. Finally, the photos are analyzed using Google’s AutoML Vision, which has been "taught" what Waldo looks like thanks to a database of images. If a match is found with at least 95% certainty, the uArm points to Waldo on the page, thanks to a weird silicon hand stuck on the end of it. You can see how it all works in the video below.
The whole process, which averages about four-and-a-half seconds with each new page, is due to Google’s Cloud AutoML. The software allows anyone to develop AI tools without knowing how to code.
Matt Reed, Redpepper's Creative Technologist, told The Verge that he built a database of roughly 130 Waldo images via Google's Image Search. "I thought that wouldn’t be enough data to build a strong model but it gives surprisingly good predictions on Waldos that weren’t in the original training set."
While it lacks any practical applications, (besides finding Waldo), it is an impressive display of what AutoML is capable of. Reed believes the software could be used to detect comic book forgeries or find what cartoon character you most resemble.
In the meantime, the rest of us will have to stick to Where's Waldo the old-fashioned way: staring endlessly at a giant splash page of tiny cartoons until our eyes are strained.