3D laser scanning for log grading: knots in intensity data

Sawmill recovery is decided by where the knots are. A grading system that finds every surface knot before the log reaches the headsaw lets the optimizer place the opening cut around them, and dark, rough, bark-on wood is exactly where ordinary area cameras struggle. A 3D laser triangulation profiler takes a different route: it projects a laser line across the log and reads both the position of that line (height) and its brightness (intensity) in one encoder-synchronized shot. You leave with the profile-rate arithmetic, the encoder settings and the ring layout for your line.

What does a laser triangulation profiler give you on a log?

Each profile is two measurements made at once: the position of the laser line gives height, and the brightness of the reflected line gives intensity, point for point at the same indices. The sensor views the projected line from an angle, so a point sitting higher on the log shifts the line's image across the sensor; that shift is the Z value. The same pixels report how much light came back, and that is the reflectance value.

Stack the profiles as the log advances and you hold a registered pair of images: a range image and an intensity image. Wrapped around the girth they become a cylinder map in which every surface point carries height and reflectance. The geometry is the same sheet-of-light construction used for 3D robot guidance with a laser profile sensor, with the reflectance channel kept instead of discarded.

If your pipeline consumes only the point cloud, half of the acquisition is thrown away. Keeping full-resolution texture next to a cropped geometric region is an established pattern — full-frame texture images alongside a cropped 3D scan — and the same discipline applies to a log's intensity image.

A ring is built from compact laser profilers of this class:

Why do knots show up in intensity before height?

Knot wood is denser than clear wood and its grain runs across the surface instead of along it, so on pine and spruce a knot commonly reads 20-50% darker in intensity while the height change over a sound knot stays under about 1 mm. The signature also carries a halo: fibres bending around the knot darken a border several millimetres wide, enlarging the target for detection.

Height catches what intensity misses. Loose and fallen knots leave depressions and holes that jump out of geometry; sound knots barely dent it. Running both channels is not redundancy but two complementary detectors on one laser line — the same split that drives surface defect detection with a 3D laser line profiler elsewhere: contrast lives in whichever channel the defect disturbs more.

Coherent laser light adds a bonus on softwoods: it scatters along the wood cells, so the line itself traces the grain deviation around a knot, the signal sawmill scanners call the tracheid effect.

Both channels inherit the quality of the peak. A noisy or clipped peak biases the centroid, and height and brightness are read from that same peak — start from clean laser line extraction before tuning any classifier.

What profile rate do you need at your log-line speed?

At 40 m/min you need about 220 profiles per second for a 3 mm axial spacing — a figure mainstream profilers exceed by more than an order of magnitude, so you size the divider, not the sensor. Current spec sheets run 2-15 kHz at 1000-3000 points per profile.

Worked example for a breakdown line:

• Line speed: 40 m/min = 40,000 mm ÷ 60 s = 667 mm/s.
• Smallest knot that changes the grade: 15 mm across.
• Profiles wanted on that knot: 5, so the spacing target is 15 mm ÷ 5 = 3 mm.
• Required profile rate: 667 mm/s ÷ 3 mm = 222 profiles per second.

A 5 kHz profiler free-running at that speed would space profiles 667 ÷ 5000 = 0.13 mm apart, about 23 times denser than the target. You are buying headroom for faster lines, darker logs and future re-grades, not capability you use daily. For scale, board scanners commonly step about 8 mm — a third of an inch — along the length; log lines scan denser because knots, not board width, set the value.

Run your own numbers through the line-rate calculator below; profile rate and line rate obey the same equation. The practical ceiling is data, not photons — three sensors streaming full profiles saturate a link long before the laser runs out, which is why onboard data reduction raises the usable profile rate.

How do you keep profiles locked to log position?

Trigger one profile per fixed distance of travel, taken from an encoder on the transport rolls, never on a free-running clock — chain speed wanders several percent under load, and time-triggered profiles smear the map. The distortion is the same one that produces stretched point clouds from wrong scan spacing.

The arithmetic, from the drive that moves the log:

• Drive roll diameter 300 mm → circumference π × 300 = 942 mm.
• Encoder 1024 pulses per revolution → 942 ÷ 1024 = 0.92 mm per pulse.
• Divider 3 → one profile every 2.76 mm, just under the 3 mm target.

The wiring is standard quadrature into the sensor's encoder input: wiring quadrature inputs for encoder-triggered capture covers the signal pairs, and setting encoder ticks per pixel covers turning counts into millimetres.

How many profilers does a 360° log scan take?

Three profilers spaced 120° apart cover the girth with overlap to stitch; on a 400 mm log each unit sees about 420 mm of arc, which 2000 points per profile samples at roughly 0.2 mm. Four units at 90° buy more overlap and a shorter standoff on large logs. The log does not need to rotate — it advances, and the ring wraps it.

Two optical constraints size the ring. Deep bark fissures shadow the line at steep viewing angles, so triangulation angles around 20-30° balance height sensitivity against occlusion; the standoff-and-angle reasoning behind laser triangulation sensor selection and optical setup applies to a sawmill ring as anywhere else. And every unit must share one encoder and one time base, or the stitched seam drifts.

Where one wide face is enough — a large log, or a sawn face — a 2.5D line-scan system covers it in a single pass. The MV-HLVS-16K100-L60 scans a 600 mm line at 16K resolution with a 120 kHz sampling rate.

How do you read dark, wet, bark-on logs?

Raise laser power and shorten exposure until the brightest glint no longer saturates, because a clipped peak shifts the centroid and corrupts height and intensity at once. Dark, wet bark absorbs most of the line and returns the rest in spikes, so set exposure against the specular spike, not the average, and expect a fraction of points rejected as saturated.

Darkness costs signal, not contrast. A knot that reads 30% darker than clear wood on a pale log still reads about 30% darker on a dark one, provided both sit above the noise floor — that is what the higher-power laser class buys. What power cannot rescue is a peak pinned at saturation, which arrives flat white in intensity and biased in height.

Chain-driven lines also shake: vibration correction in 3D height scans earns its keep when the frame moves with every log that drops on.

When does the intensity channel lie to you?

Intensity fails wherever darkness has nothing to do with wood — charred fire-killed logs, mud, ice glaze, paint marks and blue stain all fake or mask knots — and it never sees inside the log. The fix is fusion, not replacement: require agreement between height and intensity before a knot is called, and add true colour where the surface is the product.

A line scan camera under a line scan light adds that colour channel on the same encoder cadence. An 8K mono unit with an M72 mount and a 10GigE interface, such as the LS-LCM-0099, images the log face directly.

Colour from a monochrome line is a sequencing trick covered in RGB light sequencing on a monochrome line scan camera; light the face with a white line matched to its width, for example the LS-LSC-0657, a 970 mm white line light.

Internal defects are the hard limit. Rot, checks and the knots still buried toward the pith never reach the surface; X-ray and CT see them, at a cost in shielding and complexity most mills reserve for high-value decisions. The low-cost predictor is geometry itself: knots grow toward the pith, so the surface knot census plus taper constrains where they converge — which is why a full 360° scan pays for the ring.

What should you check before ordering a log scanner?

Five checks settle most orders, and each has a number attached:

• Rate headroom. Divide the maximum profile rate by your computed requirement and hold at least 5×; dust, retries and future line-speed increases spend it.
• Encoder input. Quadrature and single-ended inputs on the unit, with a divider fine enough to land just under your spacing target.
• Intensity in the data path. Confirm the SDK exposes the reflectance image with the same indexing as the height profile, not only inside the vendor viewer.
• Optics against your worst log. Triangulation angle versus bark fissure depth, and standoff clearance for branch stubs and flare; the budget behind laser triangulation accuracy specs and error sources lists what to allow for.
• Laser class and guarding. Class, interlock provisions and an enclosure rating suited to the wet debarking zone, priced into the station rather than added after.

Then commit by case: bark-on breakdown line, three profilers at 120° divided to about 3 mm; debarked grading, the same ring plus a colour line-scan face; internal rot screening, X-ray where the budget allows. Start from the 3D sensors category or the camera finder, set the line up in the configurator, and bring your log speeds and knot sizes to the applications desk — the ring geometry follows from those two numbers.