How it works · Path C

Capturing a performance

Splats can only be as good as the pictures they learn from. Capturing a performer in motion takes many angles, precise timing, and sharp frames.

3 interactive demos · about 6 minutes

Angles

A ring of cameras

To figure out where something is in 3D, the computer needs to see it from several directions. One camera can tell you a hand is somewhere along a line of sight. A second camera at a different angle tells you where on that line it actually is.

That’s why every Mirage performance is filmed by a ring of 19 synchronized 4K cameras, all aimed at the performer and recording at 60 frames per second. Together they see the front, back, and sides at every moment.

Each camera you add removes blind spots. Anything that stays hidden, like a hand behind a prop or tucked behind the body, is something the computer has to guess at. Try lowering the number of cameras in the demo and watch the gaps open up.

Try it

Why a ring of cameras

Cameras

Seen by 3+

unseen1 camera2 cameras3 or more

Number of cameras19Performer turnsAdd a chair

This is a view from above. Each dot on the performer is colored by how many cameras can see it, and a point needs several cameras to pin down where it is in space. With only a few cameras, the sides and arms fall into blind spots, and a prop like a chair blocks even more.

Timing

Every camera, same instant

Nineteen cameras means nineteen separate machines, each with its own clock. If one camera takes its picture even a few thousandths of a second after another, a fast-moving hand is in a different spot in each image, and the computer places it somewhere it never actually was.

So the cameras’ timing has to match to within a fraction of a single video frame. We also need to know exactly where every camera is and how its lens bends the picture. Our software works all of that out from the footage itself.

Try it

Why the cameras must agree on time

Time gap between cameras10 msHand speed13 ft/s

Every point in 3D is found where the cameras’ lines of sight cross. If one camera takes its picture even a few thousandths of a second late, a fast hand has already moved, and the lines cross somewhere the hand never was (the yellow ring). A dancer’s hand can easily move 10 to 20 feet per second, so every camera has to agree on the moment to well within a single video frame.

Sharpness

Freezing the motion

Regular video is shot with a fairly slow shutter. That makes motion look smooth on a flat screen, but it also blurs anything that moves fast. A blurry photo teaches the computer a blurry splat.

We film with a very fast shutter, so every frame is sharp, even in the middle of a spin. Move the shutter slider to see how much of a difference it makes to a single frame of a fast-moving hand.

Try it

Why a fast shutter

Shutter speed1/1000 sHand speed16 ft/s

A typical video camera leaves its shutter open for about 1/50 of a second, which is long enough for a fast hand to blur across the frame. Splats can only be as sharp as the pictures they learn from, so we film with a very fast shutter that freezes the movement in every frame.

Check yourself

Three quick questions

Key ideas

3D needs several angles. A point's position is found where the cameras' lines of sight cross.
Mirage uses a ring of 19 synchronized 4K cameras recording at 60 frames per second.
The cameras have to agree on timing to within a fraction of a frame, or fast motion ends up in the wrong place.
A fast shutter keeps every frame sharp, so the splats can be sharp too.

1Why can't a single camera tell you where a point is in 3D?

2A hand moves at 10 feet per second, and two cameras fire 10 milliseconds apart. How far does the hand move between their pictures?

3What does a fast shutter prevent?

0 of 3 answered