Constructing a view

The measuring point, and the step the method leaves out

Laying out equal depths correctly needs a second vanishing point that most treatments never introduce — the one belonging to the diagonals. With it the construction lands on the projected divisions to eighty femtopixels. Without it, depth is placed by judgement and the drawing depicts something nobody chose.

A perspective construction gives the directions of things easily. Run a line to a vanishing point and the direction is right, exactly, with no judgement involved.

What it does not give easily is how far back. A row of fence posts a metre apart, a tiled floor, a flight of steps, the depth of a box: all of them need the drawing to say where along a receding line a particular world distance falls, and that is the step the constructions handle least well and the treatments explain least.

There is an exact answer. It needs one extra point, and the point is not hard to find.

The measuring point, checked against the depths the camera producesFive equal depths laid out by the construction land on the projected positions to 6e-14 px.24VPcorrect from 26 cm, at 160 mm wide34° across
Fig. 1 The construction and the projection on one canvas. The marks along the ground line are measured with a ruler; the lines from them to the measuring point cut the receding edge; and the cuts land on the projected positions to 8 × 10⁻¹⁴ px over six divisions.

The idea

Along the picture plane’s intersection with the ground — the ground line — distances are undistorted. That line lies in the picture plane, so lengths on it appear at true scale, and a ruler can be used on it directly.

The problem is transferring those true lengths onto a line that recedes. The measuring point does it with one construction.

Suppose the receding direction is d, with vanishing point V. Choose a second direction m that makes some convenient angle with d — the standard choice is 45°, which makes the transfer a matter of equal lengths — and find its vanishing point M. Now:

  1. Mark the true distances along the ground line with a ruler.
  2. From each mark, draw a line to M.
  3. Where each of those lines crosses the receding line, that is where the distance falls.

M is the measuring point, and it is the vanishing point of a family of transfer lines. Nothing else is needed and nothing is judged.

Why it works

The construction is transferring a length along one direction onto another direction, using a family of parallels to do it, and it works because parallels stay parallel in the world however they look in the picture.

Concretely: take a mark on the ground line at true distance s from the origin. Draw through it the world line in direction m. That line meets the receding line at a point whose distance from the origin along d is s times a fixed factor depending only on the angle between m and d. Choose m at 45° to d and lying in the ground plane, and the factor is 1 — the transfer is exact, distance for distance.

Everything in that paragraph is a statement about the world. The picture inherits it because the projection preserves incidence: the mark, the transfer line and the receding line meet in the world, so their images meet in the picture, and the image of the meeting point is where the images meet. That is the only property being used, and it is the one thing a projection is guaranteed to keep.

Where the measuring point actually is

There is a fact about M’s position that connects the construction to the camera and is the reason the whole thing is exact.

For the standard 45° choice, the measuring point sits on the horizon at a distance from V equal to the distance from V to the station point — the eye, brought into the drawing as a point on the picture plane at the eye’s distance from it.

That distance is the focal length. So the position of the measuring point encodes the camera’s focal length, which means the construction cannot be carried out at all without committing to a viewing distance.

This is the crux, and it is why the construction gets skipped. A drawer who places two vanishing points on the paper and then places the far edges by eye has avoided ever stating how far away the viewer is. Using the measuring point forces the statement, because M’s position depends on it. Alberti’s costruzione legittima of 1435 makes this completely explicit: it draws a separate side elevation showing the eye at its actual distance from the picture plane, and reads the depths off that.

So the construction that gets left out is precisely the one that makes the viewing distance part of the drawing. Leaving it out is what allows a drawing to be made without anyone deciding where the viewer stands, which is convenient and is why so many drawings turn out to be correct only from a few centimetres away.

Where the reader has to be for a 40° picture to be correctShown 160 mm wide, this picture is a correct projection only from 22 cm away. Drawn to scale.the picture, 160 mm wide22 cm40°the eyefocal length 948 px22 cm at 160 mm wide
Fig. 2 The quantity the measuring point encodes, drawn as a plan. Choosing where M goes is choosing this distance, whether or not anyone says so.

Checking it rather than trusting it

The construction is classical, correct and easy to get subtly wrong, so the figure checks it.

The check is available because the site has both halves. The camera can be asked directly where a ground point at depth z images; the construction can be asked where it puts the division for distance z; and the two answers can be compared. Over six divisions the worst gap is 8 × 10⁻¹⁴ px, which is double-precision arithmetic and not geometry.

Building that check caught an error immediately. The first version used the vanishing point of the direction (1, 0, −1) as the measuring point, and the construction laid the divisions out in front of the ground line instead of behind it — the transfer lines ran the wrong way. The picture still looked like a receding row. The check reported a worst gap of 96,000 px.

That is the value of having the projection alongside the construction. A construction can only be checked against another construction, which is how methods with an error in them survive for centuries; a construction checked against a projection either lands on it or does not.

Diagonals, and the version that needs no measuring

There is a special case worth having, because it needs no ruler and no measuring point at all.

Given a rectangle already drawn in perspective, its diagonals cross at the image of its centre. That is an incidence, so it survives. A line through that crossing parallel to one pair of sides — meaning, in the picture, run to the appropriate vanishing point — divides the rectangle exactly in half.

Repeating gives quarters, eighths, and so on. And a second construction extends rather than divides: from the midpoint of the far side of a rectangle, a line through the far corner of it meets the extended base at the far corner of the next equal rectangle. That is how a receding row of equal bays is drawn with nothing but a straightedge, and it is exact.

The diagonal method has one limitation that explains why the measuring point still earns its place. It divides and duplicates by powers of two, and it duplicates the rectangle it is given. It cannot lay out a distance of 1.37 metres. Where the depths are arbitrary, the measuring point is the general answer and the diagonal method is the special one.

Both are exact, both are constructions in incidences only, and both are what the by-eye methods are approximating badly.

Two versions of the same invariant, one of which measures nothingFour consecutive divisions give the equal-steps method a perfect score. Using the vanishing point as the fourth point rejects it by 14%.error against the value the projection must producefour divisionsthree plus the VPthe projectionexactexactequal stepsexact14%halving3.6%25%tapering0.9%21%green: agrees with the projectiona necessary condition is not a test
Fig. 3 What the by-eye alternatives cost, in the language of the invariant. Note the first column: comparing four consecutive divisions gives the equal-steps method a perfect score, which is why the check has to bring the vanishing point in as its fourth point.

Why the ground line has to be there

One structural point about the construction, because it is what makes the whole thing possible and is easy to lose.

The construction needs somewhere in the picture where true distances can be marked with a ruler. That place is the intersection of the picture plane with the ground plane, and it is in the picture plane, so it is undistorted.

If the drawing does not include that line — if the near edge of the scene is well behind the picture plane, as it usually is — the ground line still exists and is still where the ruler goes; it is simply outside the drawn area, or coincides with the bottom edge of the sheet by convention.

This is also why the picture plane is drawn in the plan view of a classical setup. It is not decoration. It is the surface on which the ruler works, and the whole construction is a device for carrying measurements off it into the depth of the scene.

The relation to a modern workflow

Nothing in a contemporary pipeline uses a measuring point, and the reason is that nothing needs to.

A modelling program is given the world coordinates. The projection places every vertex, and the question of where a metre falls along a receding edge never arises, because the metre was specified in the model. The measuring point is a device for going from a measured plan to a picture without computing the projection, and computing the projection is now free.

What survives is the checking role. Given a drawing made by someone else, the measuring-point relation says what focal length it implies, and whether its depths are consistent with it. That is the same relation the recovery uses from the other end — three vanishing points give a focal length, and here a focal length and one vanishing point give a measuring point — and the two agreeing is one more instance of the site’s habit.

The construction is also still the clearest explanation available of why depths compress the way they do. It shows the compression as the image of a uniform transfer, which is a great deal more informative than a formula, and it makes the dependence on viewing distance visible as a distance on the paper rather than as a parameter.

What to carry away

Depth in perspective is not a matter of judgement, and the belief that it is comes from a construction that is usually taught with its measuring step removed.

The full construction needs a second vanishing point — the one belonging to the transfer direction — whose position encodes the viewing distance. Placing it is the same decision as deciding where the viewer stands, and any method that avoids placing it has made that decision silently.

With it, depths land exactly. Without it, they land wherever the drawer’s hand puts them, and the drawing depicts a solid nobody chose — not badly drawn, but drawn correctly from a camera and a scene that were never specified.

Three by-eye methods for spacing a receding row, against the truthThe posts are 1.4 m apart. The nearest by-eye method misplaces one by 11.72 m; the worst by 347.80 m.horizonequal stepshalve thetaper byworst error, in metres of depthequal steps to the horizon11.72 mhalve the remaining gap225.13 mtaper by eye347.80 mcorrect from 26 cm, at 160 mm wide34° across
Fig. 4 The alternatives, measured in metres of depth error rather than in opinions about the picture. The best of the three by-eye methods misplaces a post by 3.47 m in a row that is supposed to be spaced 1.4 m.

The tiled floor, which is the whole construction in one picture

The set-piece of Renaissance perspective is a floor of square tiles receding from the picture plane, and it is worth walking through because every part of the construction appears in it and each part can be checked.

The near edge of the floor lies along the ground line, so the tile widths across it are marked with a ruler at true scale. The tiles’ side edges run away from the viewer, so they converge to the vanishing point of the depth direction. That much needs no measuring point.

The depths do. Each row of tiles is one tile-width further back, and where those rows fall is what the measuring point supplies: mark equal distances along the ground line, run them to M, and each crossing of the receding edge is a row boundary.

Then comes the check that made the construction trustworthy in the fifteenth century, and it is a good one. In a floor of square tiles, the diagonal of every tile lies along a single straight line that runs across the whole floor — because all those diagonals are parallel in the world. So the corners produced by the depth construction must be collinear along each diagonal, and if any row is misplaced the diagonal visibly kinks at it.

That is a construction checking itself with an incidence, which is exactly the vocabulary this site uses, arrived at six hundred years ago by people who had no coordinates and no way of computing a projection. The diagonal test is why the costruzione legittima could be trusted, and it is the direct ancestor of the residual reported by the least-squares vanishing point: both ask whether things that should meet at a point actually do, and treat the failure to meet as the measurement.

The same cube turned 24° — a three-point constructionNothing about the construction changed. The number of vanishing points inside any finite distance is 3, and 1 of them fall on the canvas.horizon3 vanishing points at a finite distance2564 px · 8129 px · 531 px
Fig. 5 The case that needs two measuring points rather than one, and where the two horizontal vanishing points sit once the camera is fixed.

Two measuring points, and the case that gets skipped

Everything above concerns one receding direction. In a two-point drawing there are two, and each needs its own measuring point.

Their positions are not free. If the two horizontal directions are perpendicular in the world, then the two vanishing points V₁ and V₂ and the principal point satisfy the relation that gives the focal length, and each measuring point sits on the horizon at a distance from its vanishing point equal to the distance from that vanishing point to the station point. Both are therefore determined once the focal length is — there is nothing left to choose.

This is where the classical method and the by-eye method part company most sharply. Constructing both measuring points properly forces the drawer to fix a viewing distance and then accept every depth that follows on both directions. Placing the far edges by eye leaves both directions unconstrained independently, which is why the difference between the two placements is what decides the depicted solid rather than either placement on its own.

Put the other way round: the two-point construction with both measuring points has zero free parameters after the camera is chosen. Without them it has two, and their difference is invisible in the drawing.

What the by-eye step actually decidesSymmetric placement gives a cube for free. Eight points of asymmetry — invisible in the drawing — gives a box of side ratio 0.72, and ±16 points spans 0.52 to 1.94.0.50011.502-10010difference between the two by-eye placements (points)side ratio of the box the drawing depicts (1 = a cube)a cubeeven-handedthe one free choice in the taught methodand it decides the whole solid
Fig. 6 What the two free parameters are worth when the measuring points are not used. Symmetric placement gives a cube for free; the asymmetry decides everything else.

Doing it without the vanishing point on the paper

A last practical note, because it removes the commonest reason for abandoning the construction.

The measuring point for a 45° transfer sits at the distance from the vanishing point to the station point, and for any reasonable focal length that is a long way — usually off the sheet, often off the table. The standard response is to move the vanishing points closer together until everything fits, which quietly widens the lens and produces the exaggerated look that makes a drawing read as wrong without anyone being able to say why.

The alternative is to choose a different transfer direction. Nothing requires 45°; any angle works, with the transfer scaled by a factor that depends on it. Choosing a shallower angle brings M in from infinity toward V, at the cost of the transfer no longer being distance-for-distance and needing a multiplication.

For hand drawing that trade is usually worth taking, and it is the reason older manuals give tables of measuring points for various angles. For anything computed it is irrelevant, because a point at x = 4,300 on a 690-pixel canvas is not a problem — it is just a number, and the line to it is drawn by direction rather than by reaching it.