Direct incident light in machine vision: when it works and fails

Vertical illumination is the first technique most teams try: point a ring of light straight down the lens axis and shoot. On matt, scattering parts it is the simplest, brightest set-up you can build; on polished metal it produces a white blob and an unreadable image, often on parts from the same supplier. This guide gives you the decision rule: which surfaces direct incident light suits, how many degrees of tilt steer a glint out of the lens, what shadow that tilt costs on raised features, and what a polarizer pair costs in exposure. Every number is worked through one geometry — a 40 mm field at 100 mm working distance — so you can rescale it to yours.

What is direct incident light?

Direct incident light is illumination aimed down at the part from the camera side, arriving within roughly 15° of the optical axis, so the source, the part and the lens share nearly one line of sight.

Direct describes the arrival angles: the light reaches the part over a narrow cone instead of a broad diffusing panel, so it casts crisp shadows and delivers high intensity. A diffused ring, a coaxial light or a spot all qualify as long as their emission reaches the part steeply; the difference between a focused and a diffused source is worked through in LED lens beam shaping.

Because the emitting surface wraps around the lens path, light that scatters straight back up has almost no sideways distance to travel, so little intensity is wasted — that efficiency is the whole appeal. A ring light mounted around the lens is the standard form for fields from a few millimetres to a couple of hundred millimetres, and the trade-offs between camera, light and part positions are worked through in lighting geometry for image contrast.

Which surfaces suit it — and how bright is the return?

Matt, porous, fibrous, cast and bead-blasted surfaces suit direct incident light, because they scatter the incoming light back across the whole upper hemisphere and the lens always sits inside that scatter.

Paper, gaskets, unglazed ceramics, sintered and printed parts, food and textiles inspect well this way. The return is dimmer than engineers expect, though, because a lens collects only the light leaving the part inside its pupil cone. For the geometry used throughout this guide — a 16 mm lens at f/2.8, working distance 100 mm — the pupil is 16/2.8 = 5.7 mm across, a half-angle of atan(2.86/100) = 1.6°, and a matt surface sends sin²(1.6°) ≈ 0.0008 — roughly 0.08% — of its scattered flux into the lens.

Direct incident light still wins on these parts because every watt the LED emits is aimed at the part, with nothing spent on grazing angles or a wide diffusing panel. Run the ring at a healthy fraction of its rated current and exposures stay in the low milliseconds between f/2.8 and f/5.6.

Why does a glossy part fail under vertical light?

A flat, glossy face mirrors the source straight back up the axis: the law of reflection returns light at twice the angle it arrived, so vertical in means vertical out — a saturated blob exactly where the lens is looking.

That is the same law that separates bright field from dark field; vendors draw the boundary between the two techniques in slightly different places, but the physics is one law. On a polished part under a vertical light the specular image of the source lands inside the entrance pupil, and the camera sees the light, not the part. Worse, the highlight wanders: fixture rock, conveyor vibration and surface waviness each steer the glint by twice their own angle, so a highlight that misses the lens at set-up can reappear when the part tips half a degree. Matt surfaces scatter instead of mirroring, which is why they never show this failure.

How much tilt clears a glint — and what shadow does it cost?

Tilt the light until every glint direction misses the entrance pupil: the reflection leaves at twice the tilt angle, so the tilt must beat the worst surface slope the part can present plus a degree or two — 10° covers a production part whose worst slope reaches about 6°, with margin over the 7° floor.

Why slope, not just the pupil: for an optically flat part clamped dead level, the pupil cone is so narrow that under 1° of tilt clears it — β/2 ≈ 0.8° in the example geometry. Production parts are not that flat. Fixture tilt and surface waviness put local slopes of several degrees on the face, and each degree steers the glint by two, so the working rule is to tilt a little more than the worst slope you will accept.

Worked example: 40 mm field at 100 mm working distance

Take a 5 MP camera with 2592 × 1944 pixels and a 1/1.8" sensor about 7.2 mm wide, a 16 mm lens at f/2.8, and a part whose fixture rock and surface waviness combine to a worst slope of 6°.

• Pupil: r = 16/(2 × 2.8) = 2.86 mm; β = atan(2.86/100) = 1.6°.
• Flat-part floor: β/2 = 0.8°. Production floor: θ_min ≈ 6° + 0.8° ≈ 7°; set 10° for margin. The worst 6° patch then glints at 2 × (10 − 6) = 8° off vertical, missing the pupil by 100 mm × tan(8°) = 14 mm against a 2.9 mm pupil radius — clean.
• Shadow: the tallest feature is a 2 mm boss. Elevation = 90° − 10° = 80°, so s = 2 mm / tan(80°) = 2/5.67 = 0.35 mm.
• Pixel scale: 40 mm / 2592 px = 15.4 µm/px, so the shadow band is 0.35/0.0154 ≈ 23 px wide.
• Lens check: m = 7.2/40 = 0.18; f = WD × m/(1 + m) = 100 × 0.18/1.18 = 15.3 mm → a 16 mm lens sits right.

The decision: 10° of tilt clears glints from any patch sloping under 6°, at the cost of a 0.35 mm (about 23 px) shadow band beside every 2 mm-tall edge. If defects must be found right up to tall edges, drop to the 7° floor — shadow 2/tan(83°) = 0.25 mm ≈ 16 px — and control the fixture better, or move the search to dark field. Run your own sensor and field through the calculator below.

How much light does a polarizer cost?

Expect roughly two stops: a scene that exposed at 1 ms and f/2.8 needs 4 ms at the same aperture, or 1 ms at f/5.6, once a polarizer on the light and an analyser on the lens are in place.

The physics works in your favour on contrast: a mirror reflection keeps its polarization, so the crossed analyser blocks it, while the diffuse return keeps both components and survives at half strength — that half is the two-stop budget. The f-stop arithmetic is laid out step by step in polarizer light loss in machine vision, and the full bench set-up, including analyser rotation, in cross polarization machine vision. On a moving line the extra exposure is the real cost: if 4 ms blurs the part at your belt speed, buy the stops back with strobe current and a shorter pulse — LED overdrive and current control covers that trade.

Which light fits a direct incident set-up?

A ring light around the lens is the default for direct incident work — size its outer diameter at 2-3× the field of view so it clears the lens barrel and the field cone, which puts the 40 mm example field on a 90-120 mm ring.

The clearance check: the cone over a 40 mm field at 100 mm has a radius of 20 mm at the part; add a lens barrel of roughly 30 mm diameter and the ring's inner edge must sit beyond a 35 mm radius, i.e. 70 mm diameter, before mounting allowance. Use a 0° diffused ring for true vertical illumination; rings specified at 30° or 60° are already low-angle tools and belong to dark-field work. Under about 20 mm of field, a small spot light beside the lens, tilted the same few degrees, mounts more easily than a ring; above roughly 200 mm, tile two or more lights rather than stretch one. The shop's light finder filters rings by outer diameter and emission angle.

All three lights below run on 24 V, so a standard controller channel drives any of them:

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

The LS-RNG-0517 is a 120 mm diffused ring that fires at 0°, the textbook vertical set-up. The LS-RNG-0143 trades diameter for a built-in 30° emission angle and sits halfway between vertical and low angle. The LS-SPT-0162 is a 24 mm white spot for fields only a few millimetres wide, where a ring cannot physically fit around the lens.

When should you switch away from vertical light?

Switch when geometry, not brightness, defeats you: every 2 mm of feature height already casts a 0.35 mm shadow band at a 10° tilt, deep bores stay black no matter the exposure, and a curved mirror face aims a fresh glint at the lens from every region of the field.

Surface scratches, stamped edges and fine dust on reflective parts are dark-field work: a low-angle ring sends its light across the face, so flat areas stay dark and any disruption lights up.

A 60° ring such as the LS-RNG-0100 provides that geometry out of the box, with a 620 nm red output that suits a monochrome sensor. On curved, shiny parts a dome does the same job by surrounding the part with diffuse light from every direction — the options are compared in lighting shiny metal pins without glare. Silhouettes, presence checks and outer dimensions belong behind the part on a backlight instead. When the choice between dome, dark field and backlight had to be argued for a printed part, the inkjet OCV lighting set-up walked through it.

The decision, one line per case: matt part — vertical direct light, run bright. Glossy flat part sloping under 6° — tilt 10° and accept the shadow bands. Glossy flat part with defects up to tall edges — crossed polarizers, if the motion budget absorbs two stops. Scratches, curves or bores — dark field, dome or backlight.

What do you check before ordering?

Confirm four numbers before you order: the clearing tilt from your worst part slope, the shadow length your tallest feature casts at that tilt, the ring diameter that clears your lens cone, and the current your light draws against the controller channel that will drive it.

• Finish, on the real part. Shine a torch across it at the proposed set-up angle: a matt, even return means direct incident light works; a bright spot that moves means tilt or polarize.
• Tilt and shadow budget. Work θ_min ≈ δ_max + β/2 and s = h/tan(90° − θ) from your own field, working distance and tallest feature, as in the worked example above.
• Ring clearance. Keep the ring's inner edge beyond the field cone plus the lens barrel — a 35 mm radius, i.e. 70 mm minimum diameter, for the example geometry.
• Drive and wiring. The catalog's ring and spot lights run on 24 V; match connector and current to a controller channel, and use matched light extension cables so a pair of lights fires and dims together. On a moving line, strobe rather than run continuous — why use a machine vision lighting controller sets out the reasons.
• Comparison test. Judge set-ups on a machine vision reference target so brightness differences are measured, not remembered.

Direct incident light stays the fastest lighting to specify and the simplest to mount on a matt part, and the tilt rule above is all it takes to judge a glossy one before any hardware arrives. If you want a second opinion on the part in front of you, send a photo and your field size through the contact page, or work the choice yourself in the lighting configurator; the lighting topic page collects the other technique guides.