What Is Projective Geometry?

Look down a set of railroad tracks. In real life, the two rails never touch. They're parallel, always the same distance apart, forever.

But look at a photo of those same tracks. The rails clearly lean toward each other. Somewhere near the horizon, they meet.

So which is it? Do parallel lines meet, or don't they?

Both. It depends entirely on where you're standing. And once we take that seriously, once we stop treating "the picture" as a lie and start treating it as its own kind of geometry, we get something called projective geometry.

What does a straight line look like from a camera's point of view?

A camera doesn't see the world the way a ruler measures it. A ruler cares about distance. A camera cares about direction, about which way light travels to reach the lens.

Two rails that never meet still send light toward your eye along directions that do converge. Far away, the angle between "the direction to the left rail" and "the direction to the right rail" shrinks toward zero. In the limit, at the horizon, those two directions become one direction. The rails meet.

Drag the slider to change which way you're looking down the tracks. Notice something: however you tilt your view, the two rails always meet at exactly one point. Never two points. Never "almost meeting but staying parallel." One point, every time.

That point is called a vanishing point, and it slides around as your viewing angle changes... but it never disappears.

So what actually survives a projection?

Here's the uncomfortable part. If two parallel lines can "meet" in a picture, then a picture doesn't preserve parallelism. It doesn't preserve length either. Ordinary Euclidean geometry, the one with rulers and protractors, cares about exactly those things: lengths, angles, whether lines are parallel.

Projective geometry throws all of that out. What's left?

Incidence. Whether a point lies on a line. That's it. If three points sit on a line before you photograph them, they still sit on a line after. Straightness survives. Distance doesn't.

But here's the interesting part: there's one number, built out of distances, that does survive. Take four points $A, B, C, D$ sitting on a line. Project them from some center point $O$ onto a second line, so they land at new positions $A', B', C', D'$. The individual distances between them change completely. But this ratio of ratios doesn't:

It's called the cross ratio, and it comes out to the same number before and after the projection. Every time.

Drag the center of projection, $O$, anywhere you like. Watch the second line's points scramble into completely different spacing. And watch the cross ratio on the right stay locked to the cross ratio on the left. That single number is the thing a projection actually preserves.

Points and lines: are they really different things?

Try this thought experiment. In ordinary geometry, "two points determine a line" is a basic fact. Now swap the words: "two lines determine a point." ...that's also true, as long as the lines aren't parallel. And in projective geometry, where parallel lines meet at infinity, it's true without exception.

This isn't a coincidence. It's a deep structural fact called duality. In the projective plane, points and lines play symmetric roles. Any true statement stays true if you swap "point" for "line" and "lies on" for "passes through."

One of the cleanest places to see this is a construction called the pole and polar, defined relative to a circle. Every point $P$ outside a circle has a matching line, its polar, the line through the two spots where tangent lines from $P$ touch the circle.

Drag $P$ around. Watch its polar line swing and shift in response, always computed from the same simple rule: for a circle of radius $r$ centered at the origin, the polar of a point $(p_x, p_y)$ is the line $p_x x + p_y y = r^2$. Push $P$ inside the circle and the tangent lines vanish, there's nothing left to touch, but the polar line is still there, quietly existing on the far side. Every point has a line. Every line has a point. Neither one is more fundamental than the other.

How do we actually write "meets at infinity" as coordinates?

We've been saying parallel lines "meet at infinity" like it's a place you could point to. It kind of is, once we change how we write coordinates.

Normally a point in the plane is $(x, y)$, two numbers. Projective geometry adds a third number, $w$, and writes points as $(x, y, w)$, called homogeneous coordinates. The rule connecting the two systems: the point $(x, y, w)$ represents the same ordinary point as $(x/w,\ y/w)$, for any $w \neq 0$.

So $(2, 4, 1)$ and $(4, 8, 2)$ and $(200, 400, 100)$ are all secretly the same point, $(2, 4)$, just written with different scaling.

Now ask: what happens as $w$ shrinks toward $0$, while $x$ and $y$ stay fixed? The ordinary point $(x/w,\ y/w)$ shoots off, faster and faster, in the direction $(x, y)$... and never actually arrives anywhere finite.

Hit play. Watch the point rocket outward as $w$ approaches zero, always along the same fixed direction, the one set by the handle. That direction is the point at infinity in that direction, written $(x : y : 0)$. It's not a typo that $w = 0$ gets excluded from the ordinary $(x/w,\ y/w)$ formula. $w = 0$ is exactly the coordinate of "infinity," and it behaves like a completely ordinary point in every other way.

This is also the missing piece from the very first section. Two parallel lines, ones that never meet in the ordinary plane, turn out to share exactly one point at infinity, the one pointing in their common direction. That's the "point" your camera photographs as the vanishing point.

Why does any of this actually matter?

This isn't just historical curiosity, though it did start with Renaissance painters in the 1400s trying to figure out how to draw realistic perspective by hand.

Every 3D graphics engine, every video game, every CGI film, represents points using homogeneous coordinates for exactly the reason above: it lets a single matrix multiplication handle rotation, scaling, and perspective projection all at once, including sending faraway points toward a vanishing point correctly.

Computer vision uses the same math to straighten a photo of a document taken at an angle, or to stitch two overlapping photos into a panorama. Both tasks come down to finding the right projective transformation, the same kind of map our vanishing-point camera and cross-ratio projection are built from, that lines one image up with another.

A 500-year-old trick painters used to fake depth on a flat canvas turned out to be exactly the geometry modern computers need to understand a photograph.

The short version

A camera doesn't preserve distance or parallelism. It only preserves straightness, which points lie on which lines. Projective geometry is the geometry built entirely around that one surviving fact. Its one true invariant under projection is the cross ratio, a ratio of ratios that stays fixed even as individual distances scramble. Points and lines turn out to be two faces of the same coin, a symmetry called duality. And the trick that makes all of it precise is adding a third coordinate, $w$, so that "points at infinity," the places where parallel lines secretly meet, become ordinary points with $w = 0$ instead of a philosophical headache.

Next time you stare down a pair of railroad tracks and watch them "meet" on the horizon, you're not looking at an illusion. You're looking at projective geometry, doing exactly what it was built to do.

All visualizations are interactive React components running entirely in your browser, using SVG. No libraries beyond React.