Real-World Pokémon Pokédex Uses OpenCV and a Raspberry Pi to Identify Objects

from hackster.io

Semi-pseudonymous YouTuber Sam Makes has put together a functional Pokédex, though owing to a lack of real-life Pokémon it detects and identifies a range of real-world objects through an on-board Raspberry Pi 4 Model B.

"For me personally," Sam explains, "Pokémon was a huge part of my life growing up — and it still is today. And from Pokémon we got one of the coolest pieces of fictional tech, which is the Pokémon. You can just point it at a Pokémon and it'll tell you exactly what it is."

Born from the idea is Sam's equivalent: A handheld gadget, styled like the Pokémon original, which is capable of recognizing a range of objects — though no Pokémon — using the OpenCV computer vision framework running on a Raspberry Pi 4 single-board computer.

Housed in a 3D-printed case, the real-world Pokédex uses a low-cost camera to process a view of the world through a machine learning model based on the SSD MobileNetV3 convolutional neural network (CNN). When an object is recognized, a sprite appears on the built-in display while a computerized voice reads out an amusing custom description themed for the games and cartoon.

"It's not perfect," Sam admits, "but I'd say this is pretty freaking awesome for making it in 27 days. [It's] pretty close. I mean, yeah, it's triple the size — but I had to fit an entire Raspberry Pi in there with a bunch of components."

Links to the parts and software used are available on the Sam Makes YouTube channel, while the source code has been published to GitHub under an unspecified open-source license.