A BLIND CAMERA that turns sound into photos using AI? You need to see it to believe it!

Cameras are associated with sight. Makes sense, really. But maker Diego Trujillo Pisanty has had other ideas, using a Raspberry Pi to build a blind camera. 

Not a camera to help the blind, to be clear, but one that is – at least when it comes to visual sensors – blind. Instead of a lens, it captures sound to make a guess at the surroundings.

The camera is built around the Raspberry Pi 3B, and can also be constructed using the Pi 4. For hackers contemplating having a go, the 3D printing guides are on Diego Trujillo Pisanty's site, and you'll require the Tensorflow system.

The project is designed to be as much like a point-and-shoot camera as possible, but produce images from sound. Point the horn, press the button, wait while it is processing, then look at the image. There is a small LCD in the body.

It works using AI training, which explains why the images it captures are in color – this isn't just an echolocation system. It also, very deliberately, makes a point about the way AI is trained, as well as the advantages to an artist of developing their own AI to focus on artistic intent. Diego says "The code is as much an artwork as the object and its resulting images."

As Diego Trujillo Pisanty goes on to explain, there is a lesson about the Artificial Neural Network (ANN) which might be a warning to us all. His ANN was trained on the sights and sounds of Mexico City, and it is intentionally biased toward the city it 'knows.' If you take the 'Camera' out of the city, it will create images that look like the Mexico City.

"For the device, everything is a city," he says. The worry, too, about the potential sterilizing effect that places where AI investment is higher will have on the rest of the world's culture is all too clear.

If you're looking for other Raspberry Pi camera projects, we recently reported on a Pi analog camera conversion project, so do check that out too.