This Robot Uses Computer Vision to Find Waldo In 4.5 Seconds

An arm-and-camera pointer uses Google AutoML to spot Waldo way faster than a kid can.

|
Aug 9 2018, 3:15pm

Now even Waldo is subject to surveillance from artificially intelligent robots.

“There’s Waldo,” a robotic arm built by creative agency redpepper, serves one purpose: To find and point at that elusive cartoon-man of mystery.

Using a Raspberry Pi-controlled arm outfitted with a Vision Camera Kit designed for facial recognition, the Waldo-hunting bot searches for matches of our boy and then sends what it finds back to Google’s AutoML Vision service, which they’ve trained for Waldo-specific purposes. According to redpepper, it finds Waldo within 4.45 seconds.

Matt Reed, creative technologist at redpepper, told the Verge that he collected the Waldo images to train the AI from a Google image search, and ended up using 62 Waldo heads and 45 Waldos with his head on his body.

“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,” Reed said.

There’s Waldo is a one-stop shop for, well, finding Waldo, but it’s a great demonstration of what machine vision can do—and perhaps a much, much simpler version of how systems like Facebook and government surveillance facial recognition softwares spot individuals in a crowd. It’s a lot easier for them to find you if you wear the same red-striped shirt every day, though.