WDR and BLC in machine vision: sensor range vs software HDR

A machined part under a single light hands you both problems at once: a specular edge that saturates to white and a shadowed bore full of noise, on the same sensor, in the same frame. This guide is for engineers specifying machine vision cameras and lighting for scenes like that. You will convert a sensor's decibel rating into the contrast ratio it actually covers, measure the same ratio on your own part, and decide between one exposure with better lighting and a two-exposure strobed capture. Every decision below comes from the sensor's printed EMVA 1288 figures and your measured scene ratio, not from a feature name like WDR or BLC.

What do WDR and BLC actually do in a machine vision camera?

WDR is the camera's ability to hold very bright and very dark regions in one usable frame, and it comes from the sensor's printed dynamic range; BLC is not a range at all, but an exposure weighting that favors a chosen region of the image.

Both acronyms come from surveillance, where the camera meters the scene itself. A WDR sensor reaches its quoted 120 dB by firing interleaved exposures inside one frame readout, or by bending the response curve, so a sunlit doorway and a dim room coexist in one image. BLC — backlight compensation, sometimes written black-light compensation — tilts that metering toward a chosen window so the subject exposes correctly against a bright background.

Industrial area scan cameras mostly leave both functions out, and that is not a gap. Exposure is manual, the light is yours to place, and every photon-budget decision belongs to you. The machine-vision versions of WDR and BLC are a dynamic-range calculation and a lighting plan, which is what the rest of this guide builds.

How many decibels of dynamic range does your scene need?

Convert the printed figure before anything else: 10^(dB/20) turns 60 dB into 1000:1, 68 dB into about 2500:1 and 72 dB into about 4000:1, and the brightest-to-darkest ratio your scene produces must fit inside the converted number.

The decibel figure is measurement, not marketing: EMVA 1288 defines dynamic range as saturation capacity divided by temporal dark noise, and sensor data sheets print both numbers. Each 20 dB multiplies the ratio by ten and each 6 dB roughly doubles it. Mind the conversion: 10^(68/20) = 2512, so 68 dB covers about 2500:1, while 4000:1 corresponds to 72 dB.

The raw ratio overstates what a measurement can use. Keep the highlight under 80% of full well so nothing clips, keep the shadow at least 5σ above the noise floor so thresholds survive, and the usable scene ratio is the third line above. A 68 dB sensor then handles 400:1 of real scene, not 2500:1.

Worked example: measure the scene ratio, then demand the decibels

Run two bench exposures on the real part and let the ratio fall out.

• Exposure 1, t1 = 1 ms: the specular edge peaks at 80% of full well.
• Exposure 2, t2 = 4000 ms: the shadowed pocket reaches 50% of full well; the edge is long clipped.

Signal per unit of light is 0.8 / 0.001 s = 800 s⁻¹ at the edge and 0.5 / 4 s = 0.125 s⁻¹ in the pocket, so the scene ratio is 800 / 0.125 = 6400:1. One frame needs 20 × log10(6400) = 76.1 dB of raw range. At measurement quality — edge under 80%, pocket above 5σ — it needs 6400 × 5 / 0.8 = 40,000:1, which is 92 dB, above the tiers tabulated above for current machine-vision sensors. That is the decision: this scene splits into two exposures or gets re-lit.

What does a BLC setting change inside the image?

BLC re-weights the camera's auto exposure toward a chosen window, so that window exposes correctly while the rest of the frame drifts toward saturation, and the pixel count in the shadows does not improve by a single photon.

The mechanism is one exposure time and one gain, with the metering window deciding what they get set for. Move the window onto the dark subject and the camera lengthens exposure until the subject sits mid-scale; the bright background, which needed a shorter time, runs to 100% and clips. Nothing is created: clipped pixels stay clipped, and shadow noise stays where the photon count put it.

Industrial practice does the same thing deliberately. Set the exposure for the region you actually measure, and let the rest of the frame take care of itself. If auto exposure stays in the loop and its cap refuses to apply, check that the limit is written at startup — the auto exposure upper limit fault is exactly that failure. And when two regions need exposures further apart than the sensor's usable ratio, no weighting reconciles them; change the light or split the capture.

Can software HDR replace a true wide dynamic range sensor?

For a static part, two exposures merged in software beat any WDR chip: each frame stays linear for measurement, the combined contrast is limited only by your exposure ratio, and the cost is one extra readout, roughly doubling the frame time.

The software path is multi-exposure HDR capture: frames at different exposure times, combined into one image. Two traps decide whether it helps you.

Motion comes first. The part travels between frames and a merged image then shows ghosted edges; on-sensor WDR avoids that by taking the exposures inside one readout, at the cost of a bent response curve. The on-sensor HDR versus multi-exposure comparison works through those artifacts. For a static part, software wins on every axis that matters for measurement.

Fusion comes second. If the merged image looks identical to the long frame, the fusion step silently did nothing; the HDR fusion check gives the test. Keep exposure, gain and aperture consistent across frames so the merge math knows the scale factor — the multi-exposure consistency rules list what must match.

The combined range multiplies: frame ratio times exposure span. A 68 dB sensor at 2500:1 per frame with an 80× step covers 2500 × 80 = 200,000:1 — about 106 dB in one merged image — while every pixel you measure stays on a linear frame.

When is one exposure with better lighting better than two frames?

One frame wins whenever a fill light at about 1% of the main light's level can pull a 3000:1 shadow down to roughly 100:1, because 100:1 fits inside even a 60 dB sensor's usable 160:1 with margin to spare.

The arithmetic: a fill that adds 1% of the main light's illumination inside the shadow turns 3000:1 into 1 / (1/3000 + 0.01) = 97:1. One dim channel buys what 30 dB of sensor specification would, and it cannot clip a highlight because it never touches the bright side of the image.

Geometry does the other half. A diffuse dome or a dark-field arrangement removes the mirror-like peak instead of exposing around it, and a polarizer near Brewster's angle cuts most single-bounce glare from the surface itself. When the re-lit scene ratio lands inside the usable column of the table above, one frame per cycle is the whole design: the second readout, the merge logic and the motion bookkeeping all disappear.

How do you build a two-exposure capture with a strobe?

Fire one short exposure with the main light and one long exposure with the fill light, keep each frame linear, and take bright-region features from the short frame and dark-region features from the long frame — at a 6400:1 scene ratio that needs 0.5 ms and 40 ms.

Sequence per cycle: the trigger fires pulse A and the short exposure reads the highlights; the controller then fires the fill light and the long exposure reads the shadows. A merged image is for humans; measurement tools run per frame. For the 6400:1 bracket on a 68 dB sensor:

• Noise floor: σ = 100% / 2500 = 0.04% of full well.
• Frame A, t1 = 0.5 ms: the edge sits at 70% of full well; the shadow lands at 70 / 6400 = 0.011% of full well, about 0.27σ, and vanishes. Frame A is the highlight frame, as intended.
• Frame B, t2 = 40 ms = 80 × t1: the shadow rises to 0.011% × 80 = 0.88% of full well, which is 22σ — an SNR near 22:1, about 27 dB, plenty for edge location. The edge saturates at 70% × 80 = 5600% of full well and is discarded.
• Merge rule: any pixel at 95% or more of full well in frame B is replaced by its frame A value, rescaled by the 80× ratio if the merged image feeds a display; tools run on the original frames.

Timing and motion decide the rest. Trigger-to-pulse latency and exposure overlap belong in the timing plan, and the camera strobe output timing guide maps the pulse, the exposure window and the timestamp. Expect the first frames after a cold start to run dark while exposure and light settle; the first-frame-dark diagnosis covers that. Between reads the part moves: at 100 mm/s with 50 ms between frame starts it advances 5 mm, so stop the part or transform one frame's coordinates into the other's. A global shutter keeps each individual frame undistorted while this happens.

Overdrive shortens the long exposure: drive the fill LED harder for a shorter pulse and the same photons arrive in less time, which shrinks the motion gap. The overdrive strobing technique is built for exactly that.

Two clean reads need a camera that triggers sharply and settles fast. The MV-CH120-11UC pairs a 12 MP global-shutter sensor with USB3, and its high-speed version hardware exists for cycles like this.

On the light side, the LS-CTL-0070 is a 4-channel, 200 W digital controller with a strobe feature: main light on one channel, fill light on another, both timed from the same trigger.

If the fill pulse must be short and bright, the LS-CTL-0054 is a constant-current controller with up to 240 W per channel across 2 channels — the headroom short pulses spend.

A static part on a bench removes the timing pressure, and a mono sensor keeps the merged image clean. A 20 MP mono model covers fine shadow detail, while a compact 2.5 MP color global-shutter camera suits the small station:

How do you check a camera before you commit it to a bright-and-dark scene?

Before you commit a camera, read the sensor's saturation capacity and temporal dark noise from its EMVA 1288 data and compute 20 × log10 of their ratio yourself, because a name with WDR in it is marketing until those two numbers say otherwise.

Then check the rest against the scene, not the brochure:

• Saturation capacity and temporal dark noise from the sensor's EMVA 1288 sheet, or measure the dynamic range with a gray step target and compare the result with the print.
• Scene ratio measured on your real part with the two-exposure bench test; the 6400:1 example above came from one afternoon of exactly this.
• Exposure and gain under manual control. If auto exposure stays in the loop, know that it can land darker than your manual gain and why — the manual versus auto exposure comparison explains the gain trap.
• Trigger input and strobe output with the pulse-to-exposure timing measured on a scope, not assumed from a data sheet block diagram.
• Global shutter if the part moves during a frame; rolling shutter is fine for static parts and widens the sensor choice.
• The merge run offline on saved frames before anything is wired to the line: masks first, fusion later.

What should you order for a bright-and-dark scene?

Order by the measured scene ratio against the usable column: 160:1 and under needs only light geometry and a 60 dB-class sensor, 400:1 to 1600:1 needs a 68-80 dB sensor in one frame, and beyond 1600:1 the two-exposure strobed build is the design, plus a strobe-capable controller.

Match the examples to it and the order writes itself:

• Scene usable ratio 160:1 or less: a 60 dB-class global-shutter camera and diffuse geometry; the camera finder filters by shutter, interface and resolution once the sensor tier is known.
• 400:1 to 1600:1 usable: one frame on a 68-80 dB sensor, with the two bench exposures as acceptance evidence.
• Beyond 1600:1 usable: the strobed two-frame build of the previous sections, on a light controller that pulses per channel; the controller finder filters by channels and strobe support.

Configure the camera side from the camera configurator and the light side from the controller configurator. Bring the part, the measured ratio and the cycle time to the contact page and we will work the exposures with you, or start from the cameras topic page for the surrounding measurement guides.