Projected Interactive Display

This one is a design, not a product. Nothing has been built yet — no prototype, no hardware. It is here because the thinking is worth showing.

A projector-and-sensor unit that turns a bare wall or an ordinary desk into a touch surface. No panel, no screen, nothing on the surface itself. You touch the projected image directly — and because the projector sits very close and heavily angled, your hand never casts a shadow across what you are touching.

Depth, not RGB

The sensing is done by a depth camera. A plain colour camera aimed at a projected surface is the hardest possible case: it sees the projected image, which wrecks hand detection, and it fundamentally cannot tell hovering from touching. Depth gives touch, hover, mid-air gestures and hand identity from one sensor, and it tolerates a surface that is not perfectly flat or perfectly clean.

Each unit runs a small Linux machine. Sensing feeds a virtual multitouch device and a virtual keyboard at the operating-system level, so the OS sees a real touchscreen and ordinary applications work unmodified. A local socket carries the richer signals — hover height, hand identity, mid-air gestures — that a standard touch event cannot express.

A projector-to-camera mapping from projected calibration dots, a stored empty-surface depth reference, and a pre-warp so a heavily angled image reads as a clean rectangle. The target is under 50 ms end to end; above that, typing and dragging feel broken.

Two form factors

The wall unit mounts high or low and throws steeply onto a bare wall. Swipes, taps and gestures work well. Fast precise typing on a large wall is the acknowledged hard case — finger-to-wall distance is noisiest exactly where it needs to be sharpest — so it leans on larger keys, detecting a tap by downward velocity rather than a distance threshold, strong visual and audio feedback, and autocomplete.

The desk unit throws down onto the desktop. The camera is close, so accuracy is high and both tapping and typing are reliable. Both units run the identical sensing stack with a different calibration profile.

What you would run on it

A projected keyboard and trackpad, as an always-on-top overlay that does not steal focus, so text lands in whatever application you were already using.

And a mockup browser: you give it a prompt, and it returns candidate designs, each with a written description and a live 3D model you can orbit with your hands. Swipe through the carousel, drag to orbit, pinch to zoom, tap to select. Because text-to-mesh is still unreliable, models are assembled from a curated component library rather than generated as raw meshes.