A six-head laser log scanner: what it can and cannot measure

The Polish paper "On-Line Laser Triangulation Scanner for Wood Logs Surface Geometry Measurement" (Siekanski et al., Sensors, 2019; Warsaw University of Technology and KSM Vision, tested for a Polish floor-board maker) is a useful reference design because it reports both what works and what limits the system. It measures logs of 250 to 500 mm diameter and up to 4000 mm long on a conveyor, to optimise cutting. All figures below are from this Polish test.

Hardware

• Six identical measuring heads around the log, each with a laser line projector and an industrial camera whose optical axes are set at 30 degrees: the camera sees the laser line shifted according to surface height (laser triangulation).
• A protective pane in front of each head against dust. The authors note that panes can cause reflections and distortion.
• All six cameras are triggered simultaneously; the log position along the conveyor comes from a laser rangefinder or from trigger interval times conveyor speed.

From image to points

The software finds the laser line in each image (brightest pixel per column or an intensity-weighted centre), filters isolated pixels, removes points that belong to the casing or conveyor using a mask recorded from empty-conveyor scans, converts pixels to 3D points and moves all points to one global coordinate system.

Two-level calibration

1. Local calibration of each head: camera intrinsics and the laser plane.
2. Global calibration that ties the six heads together, using a hexagonal artefact of 232 mm radius with a different pattern on each of its six walls.

The authors name occlusion of the artefact by the conveyor as a likely source of the largest errors, together with camera noise, pane reflections and dust.

Reported accuracy

Two lessons follow. First, a ball-bar figure alone is not the whole story: on a large tube the maximum error was almost four times larger (8.02 / 2.16 is about 3.7). Second, the cutting optimisation, not the scanning, would set the cycle time: profile processing takes milliseconds, while the cut calculation takes 10 to 35 seconds. (The second point is the portal's interpretation of the paper's figures.)

Worked example: does the cycle time fit?

The paper does not state a required throughput, so use your own. Suppose a slow breakdown line handles one large log every 60 seconds. A 10 to 35 s calculation fits inside the cycle. If a high-speed line needs a decision every 10 seconds, a calculation of up to 35 s does not fit. The paper's authors tested in a floor-board sawmill; the portal's reading is that the figures may suit slower lines for large hardwood logs but need checking on fast softwood lines. The two line times above are invented for this exercise.

What the paper does not show

• No yield or recovery figures, so no financial benefit can be derived.
• Tests on a ball-bar, one steel tube and a sample log: no statistics across many logs, no data on bark condition, frozen or wet logs, or other species.
• The sample cut calculation used one product type (170 x 31 mm planks, 4 mm kerf) with plank length equal to log length.
• The paper has internal inconsistencies: the conclusions quote 2.4 mm overall accuracy while the body gives 2.16 mm.

Questions to ask a scanner supplier

• How was calibration done, and how is occlusion handled?
• How was the stated accuracy verified: on a gauge or on real, rough logs?
• What is the angular coverage, and what part of the log is not seen?

Key points

• Laser triangulation uses a laser line and a camera at an angle (30 degrees here) to compute surface height.
• Reported uncertainty about 2.16 mm in the ball-bar test, but a maximum error of 8.02 mm on a 300 mm tube, with calibration and occlusion as likely causes.
• Profile processing under 10 ms; cut calculation 10 to 35 s, so optimisation, not scanning, would limit the cycle time.
• Coverage was 320 degrees; the conveyor hides part of the log.
• The paper gives no yield gain: a business case needs a trial on your own logs.