Encoder ticks per pixel for line scan: set the scale right

A line scan camera that follows an encoder captures one line for every fixed distance the product moves, not every tick of a clock. The result is a square image: a pixel along the belt equals a pixel across it, at any speed. Set the encoder ticks per pixel wrong and every dimension along travel is wrong with it — circles print as ellipses and gauging drifts by whole percent. This guide computes the divider from belt speed, pixel size and encoder resolution, works a complete example at 300 mm/s, and covers the exposure limit and direction handling that break an otherwise correct setting. You finish with a number to type in and a test that proves it.

The rule: one line per pixel of travel

Start from the object-side pixel size: the width of belt one sensor pixel sees. If the optics deliver 0.05 mm per pixel across the belt, the belt must advance exactly that much between lines, and the encoder must release one trigger per pixel of travel. Three formulas cover the whole job:

The ×4 is quadrature: both channels, both edges, four counts per pulse. Fix the ×4 question first, because an input set to ×1 quarters the counts and your divider with it. The divider lives in the camera or grabber configuration — usually named encoder divider, decimation or StepsPerLine — and passes one line trigger to the sensor for every N encoder counts.

Line rate is the first gate: the camera must sustain it and the link must carry it. How to set the line rate of a line scan camera works that half of the problem.

Worked example: a 300 mm/s belt at 20 lines per mm

Inputs. Belt speed 300 mm/s, from the machine data. Target scale 0.05 mm per pixel, from the defect spec: a 0.2 mm defect then spans 4 pixels. Camera: 8K. Encoder on the shelf: 1024 pulses per revolution, quadrature, on a measuring wheel with 200 mm rolling circumference. Task: dimensional gauging with a 0.5 % scale tolerance.

Rate and field of view. Lines per mm: 1 ÷ 0.05 = 20. Line rate: 300 mm/s ÷ 0.05 mm = 6000 lines/s, or 6 kHz — 8K line scan sensors commonly run to tens of kilohertz, so this leaves margin. Field of view: 8192 pixels × 0.05 mm = 409.6 mm, so one camera covers a 410 mm belt. Data load: 8192 × 6000 = 49.2 million pixels per second, which a 10GigE camera carries with room to spare.

For that width, an 8K camera on 10GigE is the right call:

The line scan camera line rate calculator runs the same arithmetic on your own numbers:

Encoder arithmetic. Pulses per mm: 1024 ÷ 200 mm = 5.12. Counts per mm at ×4: 5.12 × 4 = 20.48. Counts per line: 20.48 counts per mm ÷ 20 lines per mm = 1.024. There is the failure. A divider passes whole counts, so no setting produces exactly one line per pixel of travel. Divider 1 gives one line per count: 1 ÷ 20.48 = 0.0488 mm per line, a line rate of 300 ÷ 0.0488 = 6144 Hz, and a travel scale 2.4 % too long — five times the tolerance.

Decision. Change the encoder, not the tolerance. The pulses per revolution that divide cleanly come from:

With a 200 mm wheel and 20 lines per mm that is divider × 1000. A 2000 P/R encoder gives divider 2 exactly: 2000 ÷ 200 = 10 pulses/mm, ×4 = 40 counts/mm, and 40 ÷ 20 = 2 counts per line. Set the input to ×4, the divider to 2, and the scale is exact at every belt speed. If the job were label reading or presence checks, divider 1 on the 1024 P/R part would ship at 2.4 %; for gauging it does not.

Pick an encoder that divides cleanly

Compute counts per line first. If it is not a whole number, change the encoder or the wheel, not the tolerance. On the 200 mm wheel of the example:

Aim for 2 to 10 counts per line. One count per line is the floor, because the divider cannot follow the belt more finely than its input; past ten you buy resolution the divider throws away.

A narrower web changes the camera, not the method: at the same scale, 4096 pixels cover 4096 ÷ 20 = 204.8 mm. The MV-CL042-91GC-V2 is a 4K GigE line scan camera with 7 μm pixels and an M42 mount.

Set the divider in eight steps

1. Measure the wheel by rolling it. Ten turns on a steel rule: 10 × 200 mm = 2000 mm. A worn wheel is the most common scale error, and it hides here.
2. Mount the encoder on the shaft that follows the product — the wheel rolling on the belt, not a shaft behind a gearbox.
3. Wire channels A and B to the encoder input: differential pairs (A+/A−, B+/B−) for long or electrically noisy runs, single-ended TTL for short ones. Encoder-triggered line scan capture: wiring quadrature inputs shows the connections.
4. Set the encoder input to quadrature ×4.
5. Compute and enter the divider: counts per mm ÷ lines per mm = 40 ÷ 20 = 2 in this example.
6. Cap the exposure below the line period and strobe the light, as the next section sets out.
7. Jog the belt both ways. Forward must advance the image; a backward jog must add no lines, because quadrature direction counting holds the counter until forward motion returns.
8. Prove the scale with a known length: 100 mm × 20 lines per mm = 2000 pixels.

Exposure: the limit that compresses the image

The divider is correct in distance, but exposure runs in time, and the camera cannot start a line while the previous one is still exposing. At an 18 kHz line rate the period is 1 ÷ 18,000 s = 55.6 µs — the entire exposure budget, and the reason dark line scan images come out compressed. Let auto exposure stretch past the trigger period and the camera drops lines: objects shorten along travel and thin defects disappear from the image.

The 6 kHz line of the example buys 1 ÷ 6000 s = 167 µs, but an exposure held open that long smears a full pixel, because the belt advances 300 mm/s × 167 µs = 50 µm — one pixel width — while the shutter is open. Two ways out:

• Strobe the light. Twenty microseconds of LED light per line moves the belt 300,000 µm/s × 20 µs = 6 µm, which is 0.12 pixel of smear, at a 12 % duty cycle (20 µs × 6000 lines = 120 ms of light per second). A strobe-capable digital controller such as the LS-CTL-0070 fires the light on every line:

Lighting controllers for line scan cameras covers sizing the pulse and the channel.

• Gather more light per microsecond. Bigger pixels or TDI multiply what the sensor collects at the same short exposure: the LS-LCM-0016 is a 2K line scan camera with 7 μm pixels, a 58 kHz line rate and TDI. Pixel binning on a line scan camera buys similar headroom by trading cross-belt resolution.

Either way, set a hard ceiling so the software can never drift past the line period — cap the maximum auto exposure time to stop motion blur — and keep exposure time versus shutter speed straight while you do it.

Rotary wheel or linear scale, differential or TTL

Two encoder families do this job. A rotary encoder on a measuring wheel is the default: mount it where the wheel rolls on the belt or web. A linear encoder — a read head over an adhesive magnetic tape — resolves down to 0.001 mm, which at one line per pixel of travel gives 50 counts per line, divider 50. Choose the tape when a wheel would slip, mark, or wear: it measures the web itself, and the scale doubles as a length gauge.

Both families come as differential (A+/A−, B+/B− pairs) or single-ended TTL outputs; match the output to the encoder input on your line scan cameras or frame grabbers. Rate is rarely the constraint at conveyor speeds — 40 counts/mm at 300 mm/s is 12,000 counts per second — but at high shaft speeds dropped counts appear: why encoder-triggered image capture drops images at high RPM names the causes, and Camera Link frame grabber compatibility for line scan covers the grabber side of that link.

Checks before you release it

• Known length at both ends of travel: 100 mm must span 2000 pixels near the camera and at the far end. A difference between the two means the wheel slips or the circumference is wrong.
• Direction: jog back 20 mm, then forward; the image must show no duplicated rows.
• At speed: capture at 300 mm/s and re-measure the same gauge. The divider corrects for speed by construction, so the number must not move.
• Residual error: line scan camera calibration with a moving checkerboard target removes the last fractions of a percent from lens distortion and trigger timing.

When the usual answer is wrong

• Varying speed without an encoder. Free-running at a fixed line rate distorts the image by the full speed change: a belt drifting 10 % stretches features 10 %. Encoder sync removes this by construction, because triggers follow distance, not time. If hardware triggering is impossible, make the line rate follow machine speed at runtime is the software route.
• An encoder behind a gearbox. The ratio divides your counts per mm; the FAQ works a 5:1 box that starves the divider to 0.4 counts per line. Mount on the wheel shaft and the problem disappears.
• Belt bounce on stopping. A belt that rolls back a few millimetres duplicates rows on a single-channel trigger. Quadrature direction counting holds the counter through the bounce, so buy the camera or grabber input that does it.
• Tolerance mismatch. Presence checks and code reading tolerate roughly 2 % of scale error; dimensional gauging does not. The 1024 P/R encoder ships in the first job and fails the second.

Next step: put your belt speed, pixel size and encoder pitch through the camera finder or the system configurator. If the divider refuses to come out whole, a programmable encoder or a different wheel circumference ends the hunt — bring the numbers to the contact page and we will work them with you.