Photometric stereo ring lighting for surface inspection

Photometric stereo ring lighting is useful when your inspection must separate dents, raised marks or scratches from printed patterns. The camera records the same surface under several lighting directions, then software estimates surface orientation and reflectance separately. The practical challenge is acquiring a consistent image set, not simply fitting a ring around a lens. This guide explains the geometry, sequencing and motion checks needed before selecting a light and controller.

Decide whether directional imaging fits the defect

Use independently illuminated sectors when surface relief changes the image more reliably than color or brightness alone. A shallow indentation may reverse its shading between opposite illumination directions. A printed patch on an otherwise flat, matte surface usually changes reflectance without producing that same directional shading pattern.

Photometric stereo estimates surface normals: the directions perpendicular to the local surface. It can also estimate a reflectance image under its assumed reflection model. Neither output is automatically a dimensionally accurate height map.

The distinction matters when choosing an inspection method:

For defects where directional contrast alone is sufficient, compare the approach with scratch detection using lighting and dynamic thresholding. Combining directional images may solve the inspection without reconstructing surface normals.

Establish what the light and controller must actually do

A conventional ring connected through one electrical input generally produces one illumination state. Changing its brightness does not create independent lighting directions. A photometric arrangement needs separately addressable sectors, or separate lights positioned around the viewing axis.

Before treating any ring as a photometric stereo light, confirm:

• Each required sector has an independent electrical connection or documented addressing method.
• An active sector illuminates the entire inspection region, not just its nearest edge.
• The sectors produce measurably different illumination directions.
• Unselected sectors remain off during each exposure.
• The controller can deliver the required pulse sequence through documented interfaces.

The ring lights category and lighting finder help narrow mechanical and optical candidates. However, the catalog entries supplied here do not establish independently controlled sectors for these example rings. Do not infer that capability from a ring shape, a lighting angle or a diffuser.

The standard ring example is a starting point for mechanical and illumination comparisons, not a verified segmented photometric light.

The diffused ring example provides a contrasting optical configuration. Its catalog description does not establish sector addressability either.

A diffused source can reduce visible LED structure, but diffusion may also reduce directional separation. Select the optical arrangement from sample images, not from appearance alone.

Set the illumination geometry before increasing power

Place the camera approximately along the nominal surface normal when practical. Arrange the active illumination directions around that viewing axis. Four directions spaced approximately equally in azimuth provide opposing pairs and an extra measurement beyond the theoretical minimum.

For the simplest calibrated diffuse-reflection model, at least three sufficiently independent illumination directions are needed. Four exposures provide redundancy, but they do not guarantee accuracy when highlights or shadows invalidate measurements.

As a starting experiment, try illumination elevations around 20–45° above the nominal surface plane. These are trial geometry values, not universal requirements or catalog angle definitions. Lower elevations emphasize slopes but increase shadowing. Higher elevations reduce some shadows but can weaken directional contrast on shallow relief.

For a small emitter approximated as a point source:

For example, an emitter positioned 50 mm horizontally from the center and 35 mm above the surface gives:

This estimate describes the inspection geometry, not the mounting geometry of either catalog example. A broad sector has many emitting points, so one effective direction is an approximation.

Field coverage also changes the geometry. Across a 40 mm field extending toward and away from that emitter, horizontal distances range from 30 mm to 70 mm. The corresponding elevations are approximately 49° and 27°. That variation is substantial enough to challenge a single lighting vector for the entire image.

Use a smaller inspection region, increase the lighting distance, or calibrate spatially varying illumination. Increasing distance may require more optical output to preserve exposure time.

Near-grazing illumination can operate as dark field on a nominally flat surface, emphasizing scattered light from defects. That can be excellent for detection without satisfying a simple diffuse-reflection model.

Understand what the calculation assumes

A basic photometric stereo model relates intensity to surface orientation under diffuse reflection. After subtracting the black offset and compensating for calibrated source strength, an unshadowed measurement can be written as:

Here, y_i is the corrected intensity under illumination direction i. The vector l_i describes that direction, n is the unit surface normal, and rho represents diffuse reflectance within the model.

The rows of L contain the illumination directions. Solve for g using valid measurements, then separate its magnitude from its direction. Shadowed or saturated pixels cannot simply be treated as ordinary measurements.

The calculation assumes linear image intensity, a stationary surface and sufficiently known lighting. Strong specular reflections, translucent materials, interreflections and broad illumination distributions can violate those assumptions.

A flat reference helps measure brightness nonuniformity, but it does not determine every unknown lighting direction by itself. Direction calibration needs known geometry, a suitable calibration target or another validated calibration method.

Check model agreement as well as the normal image. Compare measured intensities with intensities predicted from the fitted normal. Large disagreement identifies pixels where the model is unreliable, although low disagreement alone does not prove dimensional accuracy.

Sequence the channels and hold exposure constant

The acquisition system must associate every frame with the correct active direction. A typical four-direction cycle is: illuminate one sector, expose, switch it off, then repeat for the remaining sectors.

Prefer a documented hardware timing arrangement when timing uncertainty matters. Allow for controller response, light settling, exposure start and any required guard interval. Keep illumination within the camera's valid exposure window. Rolling-shutter operation requires particular care because different rows may integrate at different times.

Use fixed exposure, gain, aperture and image processing across the directional set. Disable automatic exposure and automatic gain during acquisition. Use linear image data where available, and avoid gamma correction or changing tone curves before reconstruction.

Equal controller settings do not necessarily produce equal irradiance. Channel output, emitter geometry and distance can differ. Calibrate their relative contributions rather than forcing every direction to produce the same average image brightness on the actual part.

The light controllers category and controller finder distinguish channel count, output type and control features. Confirm electrical compatibility separately from timing capability.

The strobe-capable controller example is listed with 4 channels and Constant voltage 24 V output. That listing does not establish an autonomous sector sequence or compatibility with an unspecified segmented light.

Ask for documented trigger behavior, channel independence, pulse limits, output loading and interface details before choosing it for a directional acquisition system. Use the controller configurator once those requirements are defined.

For pulse loading, read the strobe duty-cycle and timing guide. For acquisition order, the multi-wavelength sequencing guide addresses a related integration problem. Here, direction changes between frames; wavelength should normally remain consistent.

Calculate motion across the complete image set

A short exposure freezes motion within one frame. It does not prevent displacement between directional frames. Photometric stereo usually needs corresponding pixels to represent the same surface location throughout the set.

Start with object sampling and calculate both sources of displacement:

Consider a hypothetical inspection with a 40 mm field across 2048 pixels:

At 100 mm/s with a 50 microsecond exposure:

That exposure may be acceptable for a trial targeting less than half a pixel of blur. Now suppose four exposures start at 2 ms intervals. The first and fourth starts are separated by 6 ms:

The individual images may look sharp while their unregistered combination produces false surface structure. Shorter light pulses alone do not correct this displacement.

For a half-pixel correspondence budget at the same speed:

That is a demanding acquisition interval. Stopping the part may be more practical than obtaining the required burst rate. Image registration is another possibility, but it must be validated against relief, occlusion and lighting-dependent appearance changes.

Use the moving-part exposure guide for the within-frame calculation. The resolution calculator helps choose sampling before evaluating motion:

Verify relief separation with controlled samples

Build a sample set containing flat printed regions, genuine relief defects, acceptable texture and the expected range of surface finishes. Include parts near the inspection limit, not only obvious defects.

A useful initial test has four stages:

1. Capture each direction separately on a stationary reference. Check coverage, clipping, shadows and directional separation.
2. Repeat the complete sequence without moving the part. Look for channel drift, incorrect frame labels and inconsistent first exposures.
3. Compare reflectance and orientation outputs. Printed patches should not consistently become false relief under the validated model.
4. Repeat at production speed and across allowed part positions. Compare against the stationary result before adjusting defect thresholds.

As an initial repeatability exercise, record 100 complete image sets on a stationary sample. Track a fixed region's mean intensity for each channel and the variation in the resulting orientation estimate. Set acceptance limits from the actual defect margin, rather than borrowing an arbitrary percentage.

If the first exposure differs from later ones, investigate timing before recalibrating. The first-frame exposure and lighting checks provide a focused diagnostic path.

Document exposure, gain, lens settings, sector order, trigger timing, reference images and calibration version together. A calibration file without its acquisition settings is difficult to reproduce.

Know when the usual arrangement is wrong

Highly polished surfaces can reflect individual sectors into the camera, creating highlights rather than diffuse shading. Broad diffusion may improve appearance but weaken the directional information needed by the reconstruction. Evaluate both detection reliability and model agreement.

For glossy printed components, lighting choices for plastic lid print inspection may offer a simpler route when the actual task is reading print rather than measuring relief.

Deep grooves introduce cast shadows and occlusions. Increasing illumination elevation may recover some areas, but no algorithm can infer a directly observed surface normal from measurements that contain no usable signal.

Large fields challenge near-field geometry. Curved parts may place different regions at very different illumination angles. Translucent surfaces can redistribute light beneath the surface and weaken the local-reflection assumption.

Choose another measurement approach when absolute height, hidden surfaces or rapid continuous motion dominates the requirement. Photometric stereo is strongest when directional appearance is informative and the complete acquisition can remain geometrically consistent.

Check the specification before ordering

Use this checklist to turn a successful trial into a purchase specification:

• Optical independence: document which sectors or separate lights produce each direction.
• Coverage: verify every active direction illuminates the full inspection region.
• Geometry: record light positions, part-height tolerance and calibration assumptions.
• Electrical compatibility: confirm drive mode, load, connectors and permitted pulse operation.
• Timing: specify sector order, exposure window, switching allowance and frame identification.
• Sampling and motion: record object sampling, exposure blur and first-to-last displacement.
• Validation: retain acceptable parts, limit defects and repeatability measurements.
• Output meaning: distinguish a defect score, a normal map and a validated height measurement.

Do not order a conventional ring expecting undocumented sector control. Start the lighting configuration with your field size, sample images and timing budget, or contact the applications team to confirm whether an independently addressable arrangement can meet the inspection.