A checkerboard routine written for area-scan cameras grabs the whole board in one frame and solves camera, lens and distortion together. A line scan camera has no frame: it builds the image one line at a time, paced by an encoder, so a board that stands still returns the same row forever and gives you nothing to calibrate against. The fix is standard practice on webs, conveyors and traversing gantries: bolt the checkerboard to the moving surface and let it pass under the camera. This guide covers the hardware, the encoder settings, the measurements and the arithmetic, so you finish with two scale factors — cross-track and along-track — a deskew angle, and a length readout good to about ±0.1%.
Why the target has to move
Every line scan image is a history of motion. The sensor reports one thin row of pixels; the second dimension comes from the target traveling under the camera while the trigger fires one line per unit of travel. A stationary checkerboard therefore produces a single repeated row — a horizontal streak, useful for focus and nothing else. Motion is not an obstacle to calibrate around; it is the along-track sampling clock itself. Free-run the camera to set focus and aperture, then switch to encoder triggering before any measurement. If your toolchain cannot digest images built line by line, an area scan camera in line scan mode restores square pixels and a familiar frame interface, at a cost in maximum line rate.
The two scale factors you are solving for
Area-scan calibration returns one pixel size and a distortion model. Line scan calibration returns two independent scale factors, and conflating them is the classic failure.
| Axis | Set by | Adjust with | Main error source |
|---|---|---|---|
| Cross-track, px/mm | Lens magnification and working distance | Magnification, focus, software scale | Working-distance drift, lens distortion |
| Along-track, lines/mm | Encoder counts per mm and trigger decimation | Encoder resolution, decimation, software scale | Wheel slip and wear |
Commit to a target before measuring: hold aspect within 1.000 ± 0.005 (0.5%) for metrology, ± 1% for defect detection. The line scan field of view calculation sizes the cross-track side before you print a board, and setting the line rate falls out of the same travel-per-line number.
What you need
- A camera with an encoder trigger input. The LS-LCM-0016 is a 2K GigE camera with a 58 kHz line rate and an M42 mount — a solid first bench camera from the line scan cameras range.
- Lens mount
- M42
- Interface
- GigE
- Line length
- 2K
- Pixel size
- 7 μm
- A lens matched to sensor format and magnification: pick from line scan lenses, matching image circle to sensor as the 4K to 16K lens selection guide shows.
- An encoder geared to true travel. A measuring wheel of known circumference riding the belt, or the drive shaft: 1000 counts per revolution on a 100 mm wheel gives 10 counts/mm.
- A checkerboard. Matte 10 mm squares, photo-printed and mounted on glass or a machined plate, pitch confirmed by caliper. The board is your length standard — its error is your floor.
- Light sized for the line rate. The line scan light power guide has the arithmetic.
- Processing. A vision controller or industrial computer with a camera port and an encoder input to log lines against counts.
Calibrate the camera in eight steps
- Set the cross-track geometry first. Choose magnification so the object pixel equals the resolution you need, set working distance, and focus on the moving board in free-run. Focus only affects cross-track; lock it before anything else.
- Configure encoder triggering. Fire one line per N counts, with N = round(counts_per_mm × object_pixel_mm) so travel per line lands within a few percent of the object pixel. With the 10 counts/mm encoder and an object pixel near a tenth of a millimetre, N = 1.
- Mount the board. Flat on the belt, edges aligned to travel within 0.5°, carried by the same transport you will inspect.
- Run at production speed. Encoder pacing makes the scale speed-independent, but blur and vibration are not; calibrate at the speed you will run.
- Capture one full pass of the board with margin on both sides.
- Measure cross-track. Take sub-pixel x positions of two board edges five squares apart and divide by the calipered span between them: px_per_mm = (x2 − x1) / span_mm.
- Measure along-track. Take the line indices of the same two edges: lines_per_mm = (l2 − l1) / span_mm.
- Correct, then re-capture once. Apply the aspect scale and deskew from the next section and confirm the board reads true.
Worked example: a 205 mm web at 300 mm/s
Sensor: 2048 pixels at 7 μm pitch = 14,336 μm ≈ 14.3 mm. Lens at 0.07× magnification: object pixel = 7 / 0.07 = 100 μm, so 2048 of them span 204.8 mm across the web.
The encoder above gives 10 counts/mm, so with decimation N = 10 × 0.100 = 1 the camera fires one line per count and travel per line = 0.100 mm. Line rate = 300 / 0.100 = 3,000 lines/s at that web speed — far inside what this camera class offers.
Board pass, five calipered squares:
- Cross-track: edges at x = 411.6 and x = 914.7 px → 503.1 px / 50.00 mm = 10.06 px/mm, or 99.4 μm per pixel.
- Along-track: the same edges at line 120 and line 618.4 → 498.4 lines over the same span = 9.97 lines/mm, or 0.1003 mm per line.
- Aspect = 10.06 / 9.97 = 1.009 — the image is stretched 0.9% across track. That is under 2%, so correct in software: scale the cross-track axis by 0.991 rather than touching the optics.
- Skew: the left edge of that column sits at x = 411.6 on line 120 and x = 415.9 on line 618 — 4.3 px = 0.43 mm across 50.0 mm of travel, atan(0.43 / 50.0) = 0.49°. Deskew by 0.49° in software, or re-square the camera and re-capture.
Verification: after scale and deskew, image a 100.00 mm gauge at the left edge, center and right edge. All three must read 100.0 ± 0.1 mm; anything past 100.3 mm points at perpendicularity, not scale.
The procedure scales with the hardware, not the numbers: the LS-LCM-0099 is an 8K, 10GigE mono camera with an M72 mount for wide webs, and the LS-LSL-0150 is a 40 mm line scan lens with a 60 mm image circle. Recompute the two scale factors for the new pixel count and field; the method is unchanged.
- Lens mount
- M72
- Interface
- 10GigE
- Sensor format
- Large format
- Mount
- M58
Checks before you trust the numbers
- Caliper the board itself. Photo prints can err ±0.2 mm over 200 mm — a 0.1% floor no amount of averaging removes.
- Watch wheel slip. A wrapped, preloaded measuring wheel holds slip under 0.2%; a bare wheel on a dusty belt will not. If lengths drift, conveyor encoder calibration troubleshooting walks the fault tree.
- Track one straight board edge over 200 lines. Wander under ±1 line is normal encoder jitter; more means vibration or trigger noise.
- Compare square size at the left versus right edge. A 1° camera tilt about the line axis gives a 1.75% size gradient (tan 1° = 0.0175); sizes equal within 0.5% say the camera is square.
- If the image is tiled into frames downstream, keep every tile on one scale — line scan image overlap across frame boundaries covers the tiling.
When the usual answer is wrong
- Free-run triggering. A fixed internal clock calibrates at exactly one speed; a 5% speed ripple becomes a 5% length error. Encoder pacing is the only arrangement whose scale survives speed changes — reserve free-run for focusing.
- One pixel size. Software built for area-scan checkerboards assumes square pixels. Either feed it pre-scaled images from your own acquisition, or take the area-scan-in-line-mode route described above when the toolchain is fixed and the line rate allows it.
- TDI sign errors. Time-delay integration sums consecutive lines, so the stage direction must match the motion or every image smears. Confirm the stage sign on a TDI camera before calibrating.
- Legacy signal wiring. Plants still run older line sensors whose only output is a 4-20 mA level; when the electrical side of that upgrade matters, replacing a legacy line scan camera with a 4-20 mA output maps the wiring.
Checklist before the first production run
- Board pitch calipered and the value written into the calibration record.
- Encoder counts/mm measured over a 1000 mm tape run, not just computed from wheel diameter.
- Aspect inside target: ±0.5% for metrology, ±1% for detection.
- Deskew applied and one confirming re-capture stored.
- Gauge reads true at the left, center and right edge.
- Focus, aperture and camera mount locked; any later change triggers recalibration.
- Light and exposure fixed so moving edges stay crisp at production speed.
Ready to size the whole station? The camera configurator walks camera, lens and trigger choices in one pass, or send your belt speed and web width through the contact page and we will lay out the numbers with you.
Questions
Can a line scan camera be calibrated with a stationary checkerboard?
No. A static board returns the same sensor row on every trigger, so there is no along-track information at all — you can confirm focus and cross-track projection, nothing more. The second image axis exists only when the target moves under encoder-paced triggering. Sliding the board by hand works only if an encoder on that slide paces the lines; free-running while you slide gives a speed-smeared, unusable image.
How accurate is encoder-based calibration in practice?
Plan on ±0.1% of length across the field with a calipered board, a wrapped measuring wheel and software aspect correction. A shaft or glass-scale encoder at 100 counts/mm with a machined board pushes this toward ±0.05%. Wheel slip is the usual limit at 0.1–0.2%, which is why the gauge re-check belongs in a monthly routine rather than a one-time sign-off.
What square size should the checkerboard have?
Size squares to span 50–150 pixels cross-track: at a 0.1 mm object pixel that means 5–15 mm squares, and 10 mm is the common choice. Below roughly 20 pixels per square, sub-pixel edge noise of about ±0.1 px per edge starts to eat the scale budget. Print matte, mount flat on glass or metal, and caliper the pitch — the print is the reference, not the nominal value.
Does belt speed change the calibration?
The scale, no: encoder triggering ties lines to distance, so 150 mm/s and 450 mm/s yield the same lines per mm. Blur and vibration do change with speed. Keep motion blur under half a pixel: blur = speed × exposure, so at 300 mm/s cap exposure at 167 μs (50 μm divided by that speed). Calibrate at production speed so the vibration you measure is the one you will run.
How often should line scan calibration be re-checked?
After any change to focus, aperture, lens mount or camera position — those move the cross-track scale. After wheel or belt replacement, since wheel diameter sets counts per mm; 0.2% of wear shifts lengths by the same fraction. Otherwise a monthly two-point gauge check at the edges and center catches drift in minutes. Encoder-paced scales themselves are stable; the mechanics around them are not.
Corrections
- Description now states the key figure; keyword typo fixed.