A structured-light 3D camera is several cameras in one body. Behind a single setting (most camera SDKs call it the vision engine or the capture mode) sits the choice of projected pattern sequence, the exposure bracketing wrapped around it, and the processing that turns the captured images into a point cloud. Pick the fast phase-shift sequence for a chrome part and you get holes and false edges. Pick the longest HDR sequence for cardboard and you burn cycle time for quality the part never needed. This explainer gives the decision rule (which mode for which surface finish), the pattern arithmetic behind the capture-time cost, and the checks that confirm the choice before the cell runs.
What the capture mode actually controls
Every structured-light camera projects a known pattern onto the part, photographs the deformed pattern, and solves depth from the deformation. The capture mode bundles three decisions into one setting: which patterns the projector fires, at which exposures the camera images them, and which decoder turns the images into geometry. Change the mode and all three change together.
Two pattern families do the work. Sinusoidal phase shift projects a handful of grayscale sine fringes, each shifted by a fraction of a period; per-pixel intensity ratios give the fringe phase to sub-pixel precision, and that is where the fine detail of a point cloud comes from. Binary and Gray-code stripes project stripes that halve in width until every stripe carries a unique codeword; it costs more images, but each pixel earns an absolute stripe address that survives saturation far better than a smooth phase estimate.
The capture time follows from the image count:
Fast modes are fast because they fire few images at a single exposure. Reflection-robust modes are slow because they bracket the whole sequence at several exposures and add code patterns on top. That is the whole trade: coverage on specular surfaces bought with pattern images, and pattern images bought with milliseconds.
If you are still choosing the pattern geometry itself (lines, dots or grids), that is a separate decision, covered in structured light pattern selection for 3D. And if you build your own fringe-projection rig instead of buying an integrated camera, the imaging side is an area scan camera triggered in step with the projector, for example a 20 MP USB3 model such as the MV-CS200-10UM:
- Sensor format
- 1"
- Lens mount
- C-mount
- Interface
- USB3
- Resolution
- 16-25 MP
Everything above still applies; you simply own the sequence design instead of picking it from a list.
The decision: match the mode to the surface
A diffuse surface scatters the projected light in every direction, so almost any sequence reads cleanly. A specular surface redirects it: pixels aimed back at the projector saturate, their neighbours go black, and stacked metal adds interreflections, light arriving where no pattern was projected. Pick the mode by how much of that your scene contains:
| Surface in the scene | Mode class | Capture time | Typical jobs |
|---|---|---|---|
| Diffuse, opaque: cardboard, rubber, wood, cloth, matte plastic | Fast phase-shift | fastest | parcel induction, high-rate picking |
| Slightly specular: brushed metal, glossy plastic | Fast phase-shift or the balanced default | fast | mixed logistics |
| Specular: chrome plate, cylinders, sheet and stacked metal | HDR stripe: multi-exposure phase plus Gray code | medium | bin picking, assembly, surface finishing |
| Highly specular bins, chrome against chrome | Dedicated high-specular mode | longest | factory bin picking, inspection |
| Transparent, translucent, extremely specular, mixed loads | All-material mode | long | piece picking, parcel introduction |
The mapping mirrors a published engine table from a current structured-light camera line, and it follows the optics: the more specular the scene, the more dynamic range and absolute coding the decoder needs, and the more images the sequence carries. On that same line the balanced stripe engine is the default on the current camera generation, with the fast phase engine as the previous generation's default, which tells you where to start.
The rule: run the camera's default mode first. Drop to the fast phase mode only when the cycle is short and the parts are diffuse and opaque. Step up to the high-specular mode when metal coverage drops. Go to the all-material mode when the bin can contain anything, including clear parts. Set the mode per scene, not per camera; a cell running mixed SKUs chooses for the worst surface it must read.
In our 3D sensors range, a projected structured-light camera such as the MV-DPS200P-02 is the class that exposes these mode choices, while the galvanometer stereo MV-DLS300P-04 and the ToF MV-DT01WMU handle reflectivity by different means:
- Type
- Stereo / structured light
| Specification | MV-DPS200P-02 | MV-DLS300P-04 | MV-DT01WMU |
|---|---|---|---|
| Near field of view | 180 mm × 160 mm @ 250 mm | 280 mm × 220 mm @ 400 mm | – |
| Far field of view | 420 mm × 320 mm @ 500 mm | 620 mm × 440 mm @ 800 mm | – |
| Clearance distance (CD) | 250 mm | 400 mm | – |
| Measurement range (MR) | 250 mm | 400 mm | – |
| Repeatability (Z-axis) | 0.02 mm @ 300 mm | 0.03 mm @ 600 mm | ±(10+0.5%depth1)mm |
| VDI/VDE accuracy | 0.03 mm @ 300 mm | 0.05 mm @ 600 mm | – |
| Resolution | 3200 × 1944 @ RGB image; 1632 × 1264 @ depth map | 3200 × 1944 @ RGB image; 1632 × 1080 @ depth map | Depth map: 640 × 480 @ 30 fps; RGB image: 1280 × 1024 @ 30 fps |
| Data type | Raw image (mono + color), Depth map, RGB-D image | Raw image (mono + color), Depth map, RGB-D image | Intensity image, Color image, Depth map |
| Acquisition time | 0.2 ~ 0.7 s | 0.4~0.9 s | – |
| Image output latency | 0.5 ~ 1.2 s | 0.8~1.5 s | 133 ms |
| Dimension | 126 mm × 90 mm × 50 mm | 236.3 mm × 79.8 mm × 98.5 mm | 91 mm × 26 mm × 31.2 mm |
| Weight | 0.74 kg | Approx. 1.3 kg | Approx. 70 g |
| Price | $8,829.00 · quoted | $16,579.00 · quoted | Price on request |
Values as printed in each product's data sheet; open a part for its drawing and full specifications.
If the mode question comes before the technology question, start with industrial 3D camera selection: stereo, ToF or structured light.
Why specular surfaces break the fast mode
A phase decoder assumes the intensity a pixel sees varies smoothly, and only because the projected fringe moved. Chrome breaks that twice. Direct reflection saturates the sensor where a mirror-like patch aims the projector back at the lens and returns nothing where it aims away, so both ends of the sine clip and the recovered phase is wrong: holes in the cloud, or worse, confident wrong points. Interreflections between stacked parts add light the pattern never projected and shift the phase estimate on the neighbours too.
HDR bracketing attacks the saturation: the sequence repeats at short, mid and long exposure, and each surface point decodes from the frame where it sat well exposed. Gray coding attacks the ambiguity: a saturated pixel still yields a usable binary stripe index in at least one frame, anchoring the sub-pixel phase to an absolute address. Published measurements on highly reflective mechanical parts combine exactly these two, phase shift for resolution and Gray code for robustness.
For the practical side on metal, two related guides go deeper: improving shiny metal 3D point cloud capture and bin picking reflective metal parts. The 2D equivalent of bracket fusion is covered in HDR imaging for surface inspection.
Worked example: chrome stampings in a bin on a 4 s cycle
Inputs. Parts: chrome-plated stampings, loosely stacked, so specular plus interreflections. Sequence design: a four-step phase shift (4 sine images) plus an eight-image Gray-code block, which addresses 2^8 = 256 stripes; a common split. The high-specular mode brackets that block at three exposures. Texture image: 1. Exposure per projected image: 10 ms, set by your surface brightness; put your own measured value here and every total below scales with it. Cycle budget: 4 s per pick, a typical bin-picking robot cycle.
Fast phase mode. Images = 4 phase + 1 texture = 5. Capture time = 5 × 10 ms = 50 ms. On chrome those 50 ms buy holes and false edges, so the mode is unusable here regardless of speed.
HDR stripe mode. Images = 3 brackets × (4 + 8) = 36, plus 1 texture = 37. Capture time = 37 × 10 ms = 370 ms, which is 37 ÷ 5 = 7.4× the fast mode. Against the cycle: 370 ms ÷ 4,000 ms = 9% of one pick.
Decision. Pay the 370 ms. Nine percent of the cycle is a small cost next to the alternative: a hole where a grip point should be costs a failed pick and a retry, far more than 370 ms. Double the exposure to 20 ms and the totals double to 100 ms and 740 ms; the HDR mode still takes only 18% of the cycle. Keep the fast phase mode for the diffuse SKUs on the same cell and the average holds. Below about 1 s per part with chrome surfaces, no pattern sequence gives both coverage and speed, so change the part presentation, add a viewpoint, or move height measurement to a laser-line sensor.
One more number worth knowing: on one structured-light camera line, a capture-engine software update alone removed about 100 ms per acquisition. Capture time is as much a software property as a hardware one, so update the SDK before judging any mode.
Moving parts: when the pattern sequence is the wrong tool
A multi-image sequence needs the part still for its whole duration; at 370 ms that is a long stop on a moving line. Continuous processes such as webs, extrusions and fast conveyors use the other family: a fixed laser line swept across the part, with height decoded per line. The mode question becomes a geometry question: resolution across the line comes from the sensor, resolution along travel from the encoder.
For wide webs, a line-scan 2.5D system such as the MV-HLVS-16K100-L60 covers a 600 mm emitting length at 16K resolution with a 120 kHz sampling rate:
- Type
- 2.5D inspection
Across the line, 600 mm ÷ 16,384 pixels ≈ 0.037 mm per pixel. Along travel, a web moving 2 m/s past a 120 kHz sampler spaces profiles 2,000 ÷ 120,000 ≈ 0.017 mm apart, so the along-track pitch ends up finer than the across-line pitch and line rate is rarely the limit. If the part can stop, stay with structured light. If it cannot, see line scan cameras and the 2.5D systems in the 3D sensors range. The stop-versus-on-the-fly decision for robot cells is covered in 2D or 3D camera for conveyor pick and place.
Checks that confirm the mode choice
- Capture the same scene in every mode the camera offers and count valid points on the metal, not in total; a big total can hide exactly the surfaces that need the better mode.
- Time the capture through your API at production settings. Viewer numbers run optimistic because the viewer skips processing your application pays for, a known trap in 3D camera capture time slower in the API than in the viewer.
- Lock exposure and gain before comparing modes; auto modes drift with scene brightness and confound the comparison, as camera image brightness drift explains.
- Keep bracket settings identical between HDR test captures, for the reasons in multi exposure settings consistency for HDR capture.
- Re-check stillness after stepping up a mode: a 37-image capture takes 7.4× longer than a 5-image one, so vibration a fast mode tolerates becomes smear. 3D camera mounting stillness during capture covers the mechanics.
When the usual rule is wrong
- Matte black parts break the rule from the other side. They are perfectly diffuse, so the fast phase mode is right, but they absorb so much light that exposure, not mode, is the problem. Raise exposure or projector power before touching the mode. If you specify the finish, a matte anodize beats any capture mode for glare control, and lighter matte finishes return more signal than black dye; see specifying Type III hardcoat anodize for matte black parts.
- Transparent and translucent parts: no mode recovers clean geometry through glass or frosted plastic. The all-material mode returns partial data with confidence flags, and your grasp logic must honour them. Plan picks around what is missing.
- Grazing angles on chrome cylinders stay empty even in the longest mode; that is geometry, not software. Add a second viewpoint or re-orient the part instead of paying for repeated captures.
- Fine edges: mode choice is not only about reflectivity. The longest modes return sharper edges and fewer reflection artifacts, so a measuring task on machined edges can justify one even on mildly specular parts.
Checklist: picking the capture mode
- Classify the worst surface the cell must read: diffuse, specular, highly specular, or transparent.
- Start from the camera default; drop to fast phase only for diffuse parts on a short cycle.
- Count the images, (phase + code patterns) × brackets + texture, and multiply by your exposure; keep the total under about 10% of the cycle.
- Compare modes on valid points over the metal, at locked exposure, timed through the API.
- Update the SDK before judging capture time.
- Mixed loads with clear parts: all-material mode, with confidence-aware grasping downstream.
- Parts that cannot stop: laser line or line-scan 2.5D instead of a pattern sequence.
Compare candidates in the camera finder, configure a full system with the vision system configurator, and read more on the 3D vision topic page. If your surface sits between classes, send the question through our contact page with the part finish named, and the answer comes back with a mode attached.
Questions
Which capture mode should I set as the default on a new cell?
The camera's own default: on current structured-light cameras that is the balanced stripe/HDR mode, because it holds up on mixed loads. Reserve the fast phase mode for lines that only ever see diffuse, opaque parts, and step up to the high-specular mode when metal coverage actually drops. Re-check the choice whenever the SKU mix changes, not when the camera is installed.
How much longer does the reflection-robust mode take?
It is set by image count: an HDR stripe sequence with three exposure brackets over a 12-pattern block plus texture is 37 images against 5 for a fast phase pass, so 7.4 times the images and, at equal exposure, 7.4 times the capture time. Software also moves it: one camera line gained back about 100 ms per capture from an engine update alone, so keep the SDK current before comparing.
Does the capture mode change accuracy?
Not the scale; the camera's calibration governs that. The mode changes how much of the surface returns valid points and how sharp edges come out. Phase decoding is sub-pixel in every mode; the longer modes add absolute Gray-code addresses and exposure brackets so that sub-pixel estimate survives on bright metal instead of failing there.
One bin holds both cardboard and chrome parts. Is there one mode for everything?
Yes, that is what the all-material mode is for. It brackets widest, so every finish in the bin is well exposed in at least one frame, at a capture time between the stripe and high-specular modes. You pay the extra capture time on every pick; if that hurts throughput, sort the SKUs and run per-class modes instead.
Do transparent parts ever scan cleanly in a better mode?
No. Refraction and subsurface scatter send the pattern everywhere except back along the path the decoder expects. The all-material mode returns partial data with confidence flags; use those flags and grasp from the partial geometry. For reliable geometry on clear parts you need a different method, such as silhouette capture or a temporary coating for an offline scan.
Does a longer capture sequence need anything extra from the mount?
Yes. A 37-image capture takes 7.4 times longer than a 5-image one, and every image shares one assumption: neither part nor camera has moved. Vibration a fast mode tolerates becomes smear across the whole bracket. Stiffen the bracket, isolate the stand, or gate the capture on robot standstill.