Moving part photometric stereo: synchronize image capture

Photometric stereo depends on comparing the same surface point under different lighting directions. A moving glossy part challenges both requirements: the surface shifts between exposures, and reflections can overwhelm the shading you need. This guide helps you choose between sequential and simultaneous capture, calculate a motion budget, and test whether directional diffusion preserves useful defect contrast. The objective is a repeatable inspection image, not an unsupported promise of accurate surface height on every reflective shape.

Choose the capture method before the hardware

Use sequential directional exposures when you can freeze motion within each exposure and register corresponding surface points between images. Consider simultaneous color-separated illumination when sequential displacement becomes unacceptable, but only after testing spectral separation and material color variation.

Neither approach automatically removes glare. Photometric reconstruction needs valid intensity measurements, a suitable lighting model, and sufficient directional variation. A saturated highlight contains no recoverable shading measurement.

The photometric stereo ring-lighting guide explains the underlying directional-lighting approach. Here, the additional decision is whether your acquisition method preserves correspondence during motion.

Do not treat a camera's ability to receive one trigger as proof that it can capture a complete directional sequence. The multi-image capture guide separates that acquisition requirement from ordinary single-frame triggering.

Preserve direction while reducing concentrated reflections

A small, concentrated source can produce sharp highlights that sweep across a curved surface. Increasing the source's angular extent can spread that reflection, sometimes making useful surface detail visible over more of the part. The tradeoff is weaker directional contrast.

An ordinary diffuse dome illuminates from many directions at once. That can help produce a reference image for dirt, markings, or general appearance, but it does not provide independent directional measurements by itself.

For directional photometric inspection, you need independently controlled illumination regions or separate sources whose responses remain distinguishable. Confirm that construction explicitly. A dome shape, multiple available colors, or a controller with several outputs does not establish independently addressable lighting sectors.

Start with a stationary comparison:

• Capture each proposed lighting direction separately.
• Repeat the sequence on acceptable parts with different poses and finishes.
• Include representative defects on shallow, steep, and partially occluded surfaces.
• Record where pixels saturate or fall near the noise floor.
• Compare defect contrast against normal variation caused by shape and finish.

The dome-light guide helps explain why broad illumination can improve appearance images. The shiny-metal lighting comparison provides context when reflection control matters more than directional reconstruction.

Browse dome lights through the lighting finder for diffuse-reference candidates. Do not assume those catalog listings describe a complete directional photometric assembly.

Calculate blur and inter-image displacement separately

Exposure blur and displacement between directional images are different limits. Shortening exposure can sharpen every frame while leaving a large registration problem between frames.

For approximately constant lateral motion, use these object-space estimates:

These equations assume motion approximately parallel to the image plane and reasonably constant magnification. Rotation, height changes, and perspective can produce different displacements across the part.

Consider an illustrative field width of 100 mm sampled by 2,000 pixels:

• Sampling is 100 / 2,000 = 0.05 mm/pixel.
• A 0.25 mm feature spans 0.25 / 0.05 = 5 pixels.
• At 600 mm/s, limiting blur to 0.5 pixel gives an exposure limit of 0.5 × 0.05 / 600 ≈ 42 microseconds.
• Choosing 40 microseconds gives 600 × 0.000040 / 0.05 = 0.48 pixel of estimated blur.

Five pixels across a feature is a planning assumption, not a universal detection guarantee. Lens contrast, noise, defect orientation, and processing still determine whether that feature is detectable.

Now suppose four directional images have exposure centers separated by 250 microseconds. The first-to-last interval is 3 × 250 = 750 microseconds. During that interval, the part moves 600 × 0.000750 = 0.45 mm, or 9 pixels.

That sequence has little exposure blur but substantial inter-image displacement. It also requires an acquisition capability that must be confirmed independently; these example timings are not specifications for the catalog cameras below. Use the moving-part exposure guide when setting the initial blur budget.

Synchronize the sequence and verify actual exposure

Build timing around the camera's documented exposure behavior, not around when software requests an image. A hardware-triggered sequence still needs validated delays, pulse widths, and frame associations.

For each directional image, record:

1. The trigger or sequence event that requests exposure.
2. The actual exposure interval, using documented camera signals where available.
3. The illumination pulse and its settling behavior.
4. The direction identifier associated with the returned image.
5. The encoder position or timestamp used for motion compensation.

For a global-shutter exposure, the intended illumination must overlap the integration interval consistently. For a rolling-shutter camera, different rows may integrate at different times; a short pulse can illuminate only part of the image. Any special flash window or multi-exposure mode requires explicit documentation and verification.

Use suitable measurement equipment and a photodiode when practical to compare optical output with electrical timing. An electrical trigger trace alone cannot show whether light output has settled.

A useful initial test is 100 repeated sequences at constant speed. Check frame count, direction order, brightness consistency, and first-to-last displacement. Then repeat during acceleration and at the intended maximum operating rate. This is an engineering screening test, not statistical acceptance of a production inspection.

The shared-trigger synchronization guide explains why common triggering does not remove every timing offset. For brightness changes after startup or recipe changes, the closed-loop lighting guide helps separate optical drift from acquisition timing.

Register surface points, not just the part outline

A single translation can align a flat part moving without rotation. It may not align a curved part that rolls, tilts, or changes height. Visible boundaries can also move differently from interior surface features.

Register using features that remain stable under lighting changes. Fixture references, encoder measurements, or a separately captured texture image may be more dependable than highlights. A reflection moves with illumination direction and should not be treated as a physical landmark.

Use a conservative correspondence budget tied to the smallest defect. For example, begin by targeting residual alignment below 0.25 pixel, then measure whether that is sufficient for your inspection. A sharp intensity edge can create a false defect even with a smaller mismatch.

Test registration on acceptable parts before evaluating defect enhancement. Examine residuals near edges, grooves, occlusions, and steep changes in surface slope. Exclude areas that are not visible in every required image.

Avoid judging registration only by a smooth-looking output. Interpolation can suppress both misalignment and genuine small defects. The image-subtraction artifact guide explains why residual contour patterns can survive apparently successful alignment.

If acceptable parts produce directional artifacts that grow with speed, reduce the acquisition span or improve motion control before changing the defect threshold.

Treat simultaneous color capture as a calibrated measurement

Assigning different illumination directions to different wavelength bands can remove the time separation between directions. However, the recorded color channels are not automatically independent directional images.

A simplified linear model is:

This model requires linear, unsaturated measurements and a stable mixing matrix. If two illumination bands produce nearly the same channel response, separating them can strongly amplify noise. Calibration cannot restore information that the acquisition never distinguished.

For a three-channel system, you have at most three independent spectral measurements without additional exposures or constraints. Three suitable directions can support an idealized normal estimate, but provide no redundancy for rejecting an invalid direction. A clipped channel can make that estimate unusable.

Surface color adds another problem. A colored coating may reflect one illumination band strongly and another weakly, even when surface orientation is unchanged. That changes the apparent directional balance. Test all permitted coatings, markings, contaminants, and finish variations rather than calibrating only on a neutral target.

Disable automatic exposure, white balance, and other adaptive processing during calibration. Confirm channel timing, spatial registration, and any color-filter interpolation behavior from the camera documentation. Do not assume that a global-shutter color camera provides independent, spatially identical channel measurements.

The color-comparison guide explains why stable acquisition settings matter before interpreting channel differences.

Validate defect contrast before claiming surface shape

The simplest photometric model describes a diffuse surface under known directional illumination:

Glossy curved parts often violate this model through specular reflection, shadows, interreflection, and changing visibility. A broad directional source also represents an angular distribution, not one exact lighting vector. Treat reconstructed normals as model-dependent unless the system has been calibrated and validated for the relevant geometry.

Separate the intended inspection outputs:

• A texture or appearance image supports dirt and marking checks.
• A directional-response image can emphasize local surface changes.
• A calibrated normal estimate describes orientation under stated assumptions.
• An absolute height measurement needs further geometric constraints and validation.

For the worked motion example, first capture the same representative defect while stationary. Next, run at half speed and full speed without changing processing thresholds. Compare defect response, acceptable-part variation, invalid-pixel coverage, and registration residuals.

A system that detects the defect only while stationary has not passed the moving inspection requirement. Likewise, an attractive relief image is not evidence of dimensional accuracy.

Select parts by verified function

Use catalog components to build and test individual functions. Do not assemble a directional photometric system by matching category names alone.

LS-DOM-0085 and LS-DOM-0133 are dome-light candidates for diffuse baseline testing. Their listings do not establish independently controlled directional sectors.

For directional switching, evaluate light controllers with the controller finder. Verify output type, electrical compatibility, per-channel loading, trigger behavior, switching delay, and supported pulse operation against the complete lighting assembly.

LS-CTL-0070 is listed with a strobe feature. That feature alone does not confirm the sequence timing, channel behavior, or electrical compatibility required by your application.

Compare MV-CU020-90GC and MV-CH120-11UC as global-shutter color-camera candidates. Neither listing establishes a specialized multi-exposure mode or a calibrated color-separated photometric system.

All 2 share: Sensor type CMOS, global shutter; Exposure mode Off/Once/Continuous exposure mode; Binning Supports 1 × 1, 2 × 2, 4 × 4; Decimation Supports 1 × 1, 2 × 2, 4 × 4; Lens mount C-mount; Weight Approx. 100 g (0.2 lb.); Certification CE, RoHS, KC; Mono/color Color. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Use the area scan cameras category and camera finder to narrow the acquisition choices. Confirm sequence rate, exposure limits, trigger modes, image buffering, and sustained transfer performance from the selected camera's documentation before committing to a timing design.

Know when the usual answer is wrong

Sequential capture is a poor fit when rotation or changing occlusion prevents reliable correspondence. Faster triggering cannot recover a surface point that becomes hidden between directions.

Simultaneous color separation is a poor fit when material color varies unpredictably, spectral channels overlap too strongly, or highlights saturate essential measurements. Matching average image brightness does not resolve those failures.

More diffusion is not always better. If all illumination regions produce nearly identical shading, the system loses the directional differences needed for reconstruction. Conversely, a highly directional source may create too many shadows and concentrated reflections.

Consider a stationary dwell, an additional viewing angle, or a separate texture image when those changes simplify the task. If the requirement is traceable absolute height rather than defect contrast, evaluate a measurement method suited to that requirement instead of relying on an unvalidated relief image.

Check the complete inspection before ordering

Use this checklist to turn a promising demonstration into a defined acquisition requirement:

• Defect definition: Record minimum size, location, orientation, and acceptable variation. Representative samples matter more than visually striking demonstrations.
• Motion envelope: Include maximum speed, acceleration, rotation, height variation, and fixture repeatability. Calculate blur and sequence displacement separately.
• Lighting independence: Confirm that each direction produces a distinguishable response across the required curved surfaces.
• Valid measurements: Mark saturation, deep shadow, and occlusion explicitly. Do not silently replace invalid measurements with plausible-looking values.
• Timing evidence: Save exposure and illumination measurements alongside frame identifiers. Repeat after recipe changes and restarts.
• Color separation: Validate the mixing model on every allowed material family when using simultaneous spectral capture.
• Electrical compatibility: Confirm output type, connector pinout, channel loading, pulse limits, and required protection before connecting components.
• Production verification: Test held-out acceptable and defective parts at operating speed. Report false rejects and missed defects separately.

Bring the motion budget, representative images, and required defect coverage to the lighting configurator. If directional construction or acquisition timing remains unconfirmed, contact the applications team before selecting the final light, controller, and camera combination.