Scheimpflug optics explained for tilted surface imaging

A flat surface presented at an angle to the camera focuses along one band and blurs everywhere else: the plane of focus sits parallel to the sensor, the part does not, and the two cross in a single line. The Scheimpflug condition is the fix — tilt the sensor or the lens until the object plane, lens plane and image plane meet along one common line, and the whole tilted face snaps into focus at once. What follows gives the tilt formula, the magnification limits that make a tilt practical or absurd, a fully worked example with focal length, working distance, field of view and depth of field, and the checks to run before ordering a tilt mount.

Why is a tilted surface sharp only along one line?

Because the plane of focus stays parallel to the sensor, it cuts the tilted part along one line: on a 45° bracket with a 34 mm field the near and far edges differ by 34 mm in depth, while the lens at f/4 holds about 0.5 mm.

A lens forms a sharp image of exactly one object plane onto one image plane — the two are conjugates. Fix the sensor square to the lens and the image plane is parallel to it, so the conjugate object plane is also square to the axis. A flat part tilted 45° crosses that plane in a straight line: a sharp band, with blur growing steadily toward both edges. Nothing is wrong with the lens or the focusing; the geometry has no solution with parallel planes. The Scheimpflug principle states the requirement: image plane, lens plane and object plane must intersect along one common line. Parallel sensor and lens planes can never meet, so the condition is unreachable until something is tilted.

That leaves two ways out: make the sharp band thick enough to swallow the whole part by stopping down, or tilt the image plane so it becomes conjugate to the tilted face. The next two sections put numbers on both.

How much sensor tilt does a 45° part need?

θ' = atan(M × tan θ): a 45° part needs 5.7° of sensor tilt at 0.1x, 14.0° at 0.25x, 26.6° at 0.5x and 45° at 1x — the same part at 5x demands 78.7°, which is why Scheimpflug belongs at low magnification.

θ is how far the part is rotated from facing the camera square on: 0° means the face is parallel to the sensor, 90° means edge on. θ' is the matching tilt of the sensor away from its square position, and M is the working magnification — sensor width divided by field width.

Sensor tilt needed for a tilted flat part:

Read the pattern: tilt angle scales almost linearly with magnification at small angles, then runs away. Commit to these bands — up to 15° of sensor tilt (magnification up to about 0.27x on a 45° part) is routine hardware; 15-30° needs a long-travel tilt mount and a hard look at image-circle coverage; past 30° the sensor corner leaves the light cone and the camera body fouls the lens mount. At unit magnification the angles are equal: a 45° part wants a 45° sensor, which no practical production mount delivers.

Can I stop down instead of tilting?

Almost never at useful magnification: covering 34 mm of depth at 0.25x with a 3.45 µm blur budget asks for about f/246, and diffraction erases the image long before that — at f/11 the Airy disk is already 14.7 µm against a 3.45 µm pixel.

Depth of field in object space follows the standard machine-vision estimate:

B is the acceptable blur circle on the sensor (use the pixel size) and N the f-number. At M = 0.25 with 3.45 µm pixels the arithmetic gives 0.55 mm at f/4, 1.1 mm at f/8 and 2.2 mm at f/16. The 45° face in the next section spans 34 mm of depth — a 33.8 mm field times tan 45° — so closing to f/16 buys 2.2 mm against a 34 mm problem. Solving for the f-number: 34 × 0.0625 ÷ (2 × 0.00345 × 1.25) ≈ 246.

Diffraction closes the other door. The Airy disk diameter is 2.44 × λ × N, which at 550 nm is 5.4 µm at f/4, 10.7 µm at f/8 and 14.7 µm at f/11. Past f/8 the spot is several pixels wide, and the resolution you paid for in the sensor and lens is gone.

Two more numbers frame the decision. Depth of field scales as (1 + M) / M², so it is ten times deeper at 0.25x than at 1x at the same f-number — high magnification makes stopping down doubly hopeless. And below about 2° of part tilt, tilt nothing: at f/8 the 1.1 mm of depth covers a 33.8 mm field tilted up to atan(1.1 ÷ 33.8) = 1.9°. Stop down for the last few millimetres; tilt for the tens of millimetres.

What does a full Scheimpflug setup look like in real numbers?

A 2/3" camera with 3.45 µm pixels, a 35 mm lens at M = 0.25, a 175 mm working distance, a 33.8 mm field across the 45° face, and the sensor tilted 14.0° — five numbers fix the whole optical chain.

The task: read a 30 mm laser-marked region on a bracket face presented at 45°, with placement margin.

1. Field and magnification. A 2/3" sensor of 2448 × 2048 pixels at 3.45 µm is 8.4 mm wide; a 34 mm field with margin gives M = 8.4 ÷ 34 ≈ 0.25.
2. Lens. A 35 mm focal length suits that field from a comfortable standoff — the LS-FFL-0679 is a 35 mm, 2/3", 6 MP C-mount machine vision lens, a match for this sensor class. The machine vision lenses category filters by sensor format and resolution.
3. Working distance. The thin-lens relation WD = f × (1 + M) ÷ M gives 35 × 1.25 ÷ 0.25 = 175 mm. The lens then sits 8.75 mm farther from the sensor than at infinity focus — f × M — which is extension tube territory.
4. Sensor tilt. θ' = atan(0.25 × tan 45°) = atan(0.25) = 14.0°, tilted about the axis perpendicular to the direction of depth change.
5. Resolution check. Object-space pixel size is 3.45 ÷ 0.25 = 13.8 µm — see object space pixel size from magnification — so the 33.8 × 28.3 mm field carries a full 2448 × 2048 resolved samples, ample for a 30 mm mark.

How to choose a machine vision lens focal length walks the same trade-offs when your field size or standoff is still open.

Should I tilt the sensor or the lens?

Either one works, as long as the angle between sensor plane and lens plane changes: tilt the camera body in a graduated tilt mount behind the lens, or use a lens with a built-in tilt — rotating the whole camera and lens together does nothing, because sensor and lens stay parallel.

That parallel pair is the one thing the condition forbids. In production cells most Scheimpflug setups tilt the sensor: a tiltable C-mount or F-mount adapter between lens and camera, set to θ' and locked. Tilt lenses — machine-vision macro tilts and 35 mm photographic tilt lenses — move the lens instead; their travel tops out near 8-9°, which covers a 45° part only up to M ≈ tan 8° = 0.14. That is enough for circuit boards and labels, and short of the 14-27° band larger magnifications need.

Around a telecentric lens the same tilt applies between lens and sensor, and space often gets tight; the LS-LXR-0144 double-fold prism turns a telecentric lens's output by 90° over a 30 mm optical path when the camera will not fit the cell. Hardware of this kind lives with the lens accessories.

Set the tilt axis along the direction where part depth does not change: for a part tilted left-to-right, rotate the sensor about the vertical axis. Tilting about the wrong axis leaves the blur exactly where it was.

Which lens types work in a Scheimpflug setup?

Fixed-focal, macro and telecentric lenses all work — the Scheimpflug condition is plane geometry, not a lens property — but for measuring on a tilted face telecentric is the pick, because it holds magnification constant while an entocentric lens varies by about ±10% across the same field.

The tilt goes between whatever lens you choose and the camera, so the lens itself is an ordinary one: choose it for format, resolution and working distance exactly as for a square-on setup, then add the tilt. The difference shows up in the image. On the worked example an entocentric lens runs from about 0.28x at the near edge to 0.23x at the far edge — a 20% scale change across the field — so a square grid images as a trapezoid. Presence checks, OCR and code reading tolerate that; direct millimetre measurements do not, short of per-region calibration and the methods in lens distortion correction for pixel-to-mm measurements.

A telecentric lens removes the scale change: magnification is constant across the field, so one calibration covers the tilted face. The trade-offs are the ones in entocentric versus telecentric lenses and when you need a telecentric lens for gauging, and the telecentric lenses category is filtered by magnification and working distance. The LS-TCL-1422 is a 0.166x bi-telecentric C-mount lens with a 258 mm working distance for a 1" sensor. The LS-TCL-0574 is a 0.4x fixed-magnification lens with a 227 mm working distance for a 1.2" sensor. Two low-magnification telecentric candidates for a tilted-face cell:

At 0.166x the 45° face needs atan(0.166 × 1) = 9.4° of sensor tilt, comfortable for any tilt mount. At 0.4x it needs 21.8°, the top of the practical band from the table above. Depth of field around a telecentric lens has its own behaviour on a tilted plane — telecentric depth of field validation covers how to check it.

How do I set the tilt and check it?

Compute θ', preset the tilt mount, focus at the centre of the field, then walk the focus between the two field edges until both are sharp together — and hold the tilt within about 0.2°, because a 1° error leaves 2.6 mm of edge defocus against 0.55 mm of depth of field.

The tolerance deserves its arithmetic. Through the criterion, a 1° sensor error swings the conjugate object plane by about 2.1° at M = 0.25, and tan 47.1° − tan 45° = 0.077 across a 33.8 mm field is 2.6 mm of defocus at the edges — five times the f/4 depth of field. So iterate rather than trust the graduation alone:

1. Compute θ' from the working M and preset the tilt mount to it.
2. Present the part, or a flat stand-in, at the real face angle; presentation must repeat within about 2° part to part, or the tilt chases a moving target.
3. Focus at the centre of the field, then check the two edges along the tilt direction.
4. Split the difference: nudge the tilt toward the blurrier edge, refocus, re-check. Two or three passes converge.
5. Lock focus, aperture and tilt — lens locking screws — and re-check after locking, which always moves something.

For the check itself, lay a dot grid flat on the tilted face: the LS-LXR-0029 calibration board carries 0.5 mm dots on a 76 x 76 mm grid. If the tilted face is also a datum for touch probes elsewhere in the cell, the same geometry misleads them — see probing tilted surfaces and fixing wrong datum results.

When is a Scheimpflug tilt the wrong answer?

Above about 0.5x on a 45° face — 26.6° of sensor tilt — or anywhere past 1x, where the same part wants a 45° sensor; also whenever the part angle is not repeatable, or whenever you can simply square the part up to the camera.

Work down the list in order of cost. Squaring the part or the camera is a fixture change that deletes the problem: the best Scheimpflug setup is the one you avoid. Below about 2° of tilt the depth of field already covers the field (f/8 gave 1.9° on the worked example), so stop down instead. If the face angle varies more than about 2° from part to part, a fixed tilt cannot follow it — fix the fixture, drop the magnification, or move the geometry task to a 3D sensor. And tilt fixes exactly one plane: a face with deep machined features needs focus stacking or 3D, not a tilted sensor.

Before ordering, check these off:

• Working magnification computed as sensor width ÷ field width, with the real sensor.
• θ' = atan(M × tan θ) under 15°, or a deliberate decision to live in the 15-30° band.
• Lens image circle at least the sensor diagonal divided by cos θ'.
• Tilt axis perpendicular to the direction of depth change on the part.
• Part presentation repeating within about 2°.
• Lockable focus, aperture and tilt hardware, and a dot grid to prove the result.

Start with the lens finder or the configurator above, using the field width and the part angle as inputs. If the numbers land awkwardly, send both through the contact page and the applications team will work the geometry with you.