top of page
  • amyo22

There’s Waldo!

‘Where’s Waldo’, a childhood game of many, was successfully beaten by redpepper’s AI.

How does it work?

With its metal robotic arm (containing a uArm Swift Pro and a Vision Camera Kit to detect faces), the AI took a photo of the page and employed OpenCV to pinpoint possible Waldo faces. Then, it sent the individual images to AutoML Vision by Google. This service effectively trained the images without needing users to understand coding, and has been used by many to classify images ranging from flowers to animals. Once the AI found a 95% confidence match, it moved its hand to the correct Waldo among a sea of other faces.

The leader of the project was Matt Reed, who gathered a photo collection of 45 full body Waldos and 62 Waldo heads simply from Google search, then sent this to AutoML Vision. He initially thought that the dataset was too small but, once the prototype AI recognized Waldo within 4.45 seconds, he was proven wrong.

What does this mean?

This instance of image recognition scratches the surface of what AI vision can do. It can be extended to fun applications like puzzle solving or more serious ones like detection of handwriting forgery. In fact, the image recognition market is predicted to reach 3.8 billion USD this year. Surely, more and more interesting image recognition algorithms will be implemented in our daily lives, to the point where we might not even realize it!


Written by Amanda Y


bottom of page