A shiny part can produce a convincing surface near the image center and almost no usable depth elsewhere. Raising matching sensitivity may add points without recovering the actual shape. This troubleshooting workflow is for engineers commissioning reflective-metal inspection or localization. Start with the raw captures, separate exposure problems from viewpoint-dependent reflections, and validate each change against a fixed reference. The aim is a repeatable measurement, not simply a denser point cloud.
Diagnose the raw captures before the cloud
On a stereo system, depth depends on finding corresponding surface information in separate views. A reflective highlight can move with the viewpoint instead of identifying one physical surface location. One camera may see a bright reflection while the other sees a dark area.
Inspect the original intensity images from every receiving camera, including projected-pattern captures where available. A rendered cloud can conceal the reason correspondence failed. Disable display auto-scaling when comparing images, or record its settings so brightness changes remain interpretable.
| Observation | Likely explanation | First test |
|---|---|---|
| Bright regions lose all visible pattern or texture | Saturation or excessive local contrast | Reduce exposure and inspect the original pixel values |
| A feature appears bright in one view and dark in another | Viewpoint-dependent reflection | Change part or camera angle slightly |
| Dark regions contain little usable detail | Insufficient received signal or unfavorable reflection geometry | Compare a longer exposure with a changed viewpoint |
| Holes follow edges or deep recesses | Occlusion or inadequate shared visibility | Inspect both views and change the viewing geometry |
| More points appear only after relaxing rejection settings | Ambiguous correspondences may be entering the cloud | Check depth against a known reference |
For a broader diagnostic sequence, use the guide to finding the causes of missing depth points. Here, concentrate on whether the metal provides information that stays consistent between views.
Establish a baseline and measure saturation
Hold the part, camera, processing region and acquisition settings fixed. Save the raw images, calibration identifier, settings export and resulting cloud before changing anything.
Choose an inspection region that represents the actual measurement surface. Exclude the background when calculating coverage; otherwise, a good background can hide poor performance on the metal.
For an initial diagnostic, capture 10 stationary acquisitions at the baseline settings. Record:
- Saturated-pixel fraction in each receiving image.
- Valid-depth fraction inside the inspection region.
- Variation in the measured feature across acquisitions.
- The location of persistent holes and isolated depth clusters.
Calculate saturation from original image values, not the brightness of an exported screenshot. For integer images, count values at the sensor or acquisition pipeline's documented clipping level.
That percentage is descriptive, not an acceptance limit. A small clipped area can still invalidate the exact edge or seating surface you need. Evaluate its location as well as its size.
Test geometry without assuming the center is wrong
Good depth directly beneath the camera does not contradict the physics of reflection. The receiving optics, illumination direction, surface curvature and finish determine which light reaches each camera. A favorable central patch may provide usable scattered light or a projected pattern visible to both views.
Elsewhere, curved metal can direct illumination away from one receiver or create a clipped highlight. There is no universal rule that the image center must be the worst location.
Run a controlled angle test. For example, compare the baseline orientation with approximately 5° and 10° changes, provided the documented measurement volume and mechanical clearances permit them. Change one axis first, then repeat around the other axis if necessary.
A slight tilt can move a troublesome reflection away from a measurement region. It can also create occlusion or move the surface outside the calibrated range. Check both effects rather than selecting an angle solely because its cloud looks fuller.
Use the working-distance and positioning guide when evaluating the permitted measurement volume. Keep the inspection feature visible to all receivers required for depth reconstruction.
Separate exposure changes from matching changes
Return to the baseline geometry before testing exposure. Keep matching and filtering unchanged so the optical effect remains identifiable.
For a stationary trial, an illustrative exposure bracket is 1 ms, 2 ms and 4 ms, if those settings are supported. These are test values, not recommended settings for a particular sensor. Check each receiving image for clipping, visible pattern detail and dark-region noise.
A shorter exposure may recover detail in bright areas while removing it from darker metal. A longer exposure may improve dark regions while saturating highlights. If no single exposure preserves useful information across the inspection region, evaluate the manufacturer's documented multi-exposure acquisition mode, where available.
Multiple exposures do not automatically resolve specular reflections. They address dynamic range only when at least one exposure contains valid correspondence information. A reflection that changes appearance between viewpoints remains problematic at every exposure.
For moving parts, include the entire acquisition sequence in the motion budget. Separate exposures can represent different surface positions, producing artifacts during reconstruction or fusion. The guide to HDR acquisition and fusion for inspection explains the related dynamic-range tradeoff, although intensity-image fusion and depth reconstruction are different operations.
Understand why a dense cloud can still be inaccurate
For an ideal rectified stereo pair, depth follows the relationship below. This simplified model helps explain sensitivity; it is not a substitute for the sensor's calibration.
Consider an illustrative system with a focal length of 800 pixels and an 80 mm baseline. At 500 mm depth, the expected disparity is:
These are hypothetical geometry values, not catalog specifications. The estimate concerns small correspondence errors. A false match caused by a reflection can produce a much larger error that this approximation does not describe.
Increasing point count therefore does not prove improved accuracy. Check a known distance, reference plane or dimensional feature independently. For a cloud that appears smoothly bowed, the guide to diagnosing curvature on a nominally flat surface helps distinguish shape errors from missing coverage.
Adjust correspondence rejection only after capture improves
Once the raw images contain usable detail, evaluate the manufacturer's documented matching options. Parameter names and numerical scales differ between systems, so a threshold from another camera is not a transferable recipe.
A practical test order is:
- Keep the best verified geometry and exposure fixed.
- Change one documented matching or confidence setting.
- Reprocess the same saved capture when the software permits it.
- Compare coverage, outliers and reference error together.
- Restore the baseline before testing a different parameter family.
Relaxing ambiguity rejection may admit useful points, but it can also accept incorrect correspondences. Tightening it may remove unreliable depth while leaving a visibly incomplete cloud. Neither direction is inherently better; the measurement requirement decides.
Apply smoothing and hole filling last. They can make a cloud easier to display without adding measured surface information. Keep inferred regions distinguishable from directly reconstructed points wherever the software supports that distinction.
A useful acceptance report records both valid coverage and dimensional error. For example, an increase from 70% to 90% coverage is not an improvement if the added points shift the measured feature beyond its tolerance.
Choose illumination and sensor changes deliberately
Browse 3D sensors by measurement method, then use the camera finder to identify candidates for evaluation. The catalog includes different acquisition principles; their presence in the category does not establish suitability for reflective metal.
The MV-DPS200P-02 is cataloged as a projected structured-light camera. Treat it as a candidate for a sample-based feasibility assessment, not a guaranteed cure for specular surfaces.
- Type
- Stereo / structured light
The MV-DP4180-03P is cataloged as a laser profile sensor. Profiling changes the acquisition geometry and workflow, but reflective return and visibility still need testing.
- Type
- Laser profiler
Diffuse lighting deserves a separate decision. Dome lights can reduce objectionable reflections in an intensity image by surrounding the part with diffuse illumination. That does not guarantee better active depth capture: added light can reduce projected-pattern contrast or be ineffective at the receiving wavelength.
The LS-DOM-0085 is a dome light. Its inclusion here identifies an illumination option to evaluate for an accompanying intensity inspection, rather than a verified depth-improvement accessory.
- Size
- 100-200 mm
- Voltage
- 24 V
The guide to dome, diffuse and polarized lighting for shiny pins addresses the related intensity-image problem. Do not transfer its lighting choice directly to an active depth system without testing. Use the lighting finder or light configurator after establishing which illumination geometry helps the actual task.
Verify repeatability and recognize optical limits
Repeat the baseline acquisition sequence after choosing candidate settings. Test the expected range of part positions, orientations and surface finishes, not just the easiest sample.
Measure coverage specifically where the inspection uses it. A broad, well-reconstructed top surface cannot compensate for missing depth on the narrow edge used for alignment. Follow the same principle when using the point-cloud inspection workflow for pin features: evaluate the feature, not only the overall cloud.
Check surface-condition variation separately. Residue, machining direction and finish can change the reflected signal. Keep sample preparation consistent with the production process; an unusually clean demonstration sample may not represent normal parts.
Some surfaces do not provide reliable correspondence from the permitted viewpoint. If every exposure produces conflicting reflections or persistent occlusion, more permissive matching is unlikely to create trustworthy measurements.
Possible next steps include another viewpoint, a different measurement principle, or process-approved surface preparation. A temporary matte coating can help a feasibility study only when contamination and dimensional changes are acceptable. Do not treat coated-sample performance as evidence for an uncoated production process.
Capture-to-order checklist
Before committing to settings or replacement hardware, confirm the following:
- Raw evidence saved: retain every receiving image, calibration identifier and settings export so the failure can be reproduced.
- Geometry tested: compare controlled angle changes while staying within the documented measurement volume.
- Exposure isolated: test clipping and dark-region detail without changing matching settings simultaneously.
- Reference measured: compare reconstructed geometry with an independently known feature.
- Coverage localized: report usable depth on the measurement region rather than across the whole image.
- Processing separated: distinguish reconstructed points from smoothing or hole-filled estimates.
- Production variation represented: include the expected finishes, orientations and motion conditions.
- Acceptance defined: agree on dimensional error, repeatability and permitted missing areas before selecting hardware.
For a feasibility review, prepare representative raw captures, settings, the measurement tolerance and the required cycle time. Share permitted recordings through application support before selecting a sensor or lighting change. That evidence makes the next decision about a specific optical limitation rather than another round of threshold adjustments.
Questions
Why does shiny metal sometimes reconstruct better near the image center?
The center can have favorable illumination and viewing geometry even when surrounding areas produce strong reflections. Surface curvature and finish determine what each receiver sees. Compare the raw receiving images at good and bad locations before assuming that central reflections should always make depth worse.
What exposure should I use for reflective metal?
There is no transferable exposure setting for reflective metal. Start with a stationary part and compare supported exposures while holding geometry and matching constant. Select settings that preserve usable detail in all required receiving views, then verify the reconstructed dimensions rather than judging image brightness alone.
Will a dome light improve active depth capture?
A dome can improve an accompanying intensity image, but an active depth system depends on its own illumination and reconstruction method. Additional diffuse light may reduce projected-pattern contrast or have little effect at the receiving wavelength. Test raw captures and dimensional accuracy before specifying supplemental lighting.
Should I lower matching rejection thresholds to recover missing points?
Relax a rejection setting only after the raw captures contain credible correspondence information. Evaluate the same saved capture where possible, and compare reference error alongside coverage. Extra points are useful only when they represent the surface accurately; a fuller cloud can contain more false matches.
When should I evaluate a different sensor or measurement method?
Consider another approach when controlled exposure and viewpoint tests still leave the measurement feature without consistent information. Provide representative unmodified samples and a defined tolerance for evaluation. A different viewpoint or acquisition principle needs its own feasibility test rather than an assumption that it handles every reflective finish.