Wally in grass

‘There’s Wally’, says robot dedicated to solving ‘Where’s Wally?’ puzzles

Image credit: Dreamstime

A creative agency, Redpepper, has built a robot capable of rapidly finding and pointing at the bespectacled star of the ‘Where’s Wally?’ series.

The ‘Where’s Wally?’ books are a popular series of children’s puzzle books which require the reader to locate the character of Wally – identifiable by his round glasses, candy-striped shirt and bobble hat – in humorous and densely populated illustrations. The books are published around the world under different names; in the US, Wally is known as ‘Waldo’.

Now, a team at Redpepper have built a robot, There’s Waldo, dedicated to solving the puzzles.

Its robotic arm is a uArm Swift Pro, which comes complete with a downwards-facing camera with facial-recognition capabilities. The camera stands above the open pages of the book and takes a high-resolution photograph. It then uses OpenCV – an open-source library of functions mostly for image recognition – to identify a group of possible faces.

Google’s Cloud AutoML Vision, which had been prepared with many pictures of Wally, then compares the candidates’ faces with Wally’s likeness.

If the program can find a match of at least 95 per cent confidence, it will move its arm to above the face. The middle finger of a limp, silicone hand hanging from the arm lands on the likely face (or faces, if more than one is identified as Wally). According to Redpepper, the robot – which is still just a prototype – is able to spot the character in as little as 4.45 seconds; much faster than a child searching the pages unassisted.

Google’s AutoML is a free-to-use service which allows users to train neural networks for image recognition, natural language processing and translation with minimal expertise. While facial recognition normally requires at least thousands of images to perform reliably (and even then is a source of controversy), There’s Waldo was able to recognise the character from a limited set of just dozens of images, thanks to the simplicity of the drawings.

The robot is unlikely to be of much practical use, although it is a light-hearted demonstration of what is possible with services like AutoML, which essentially allow anybody to create neural networks for their own purposes.

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them