Six laser heads, one log: what an on-line scanner can and cannot measure

Introduction

Every sawmill that breaks down hardwood or softwood logs faces the same early decision: how should a log be positioned and cut so that the greatest possible value ends up in boards? Round wood is never a perfect cylinder. Sweep, taper, ovality, butt swell and bark remnants all change what can be recovered. Human operators judge this by eye, but a camera-based system can measure the full outer surface in three dimensions and let software test many cutting layouts in seconds.

A Polish research team from Warsaw University of Technology and the company KSM Vision describes such a system in the open-access paper "On-Line Laser Triangulation Scanner for Wood Logs Surface Geometry Measurement", published in the journal Sensors in 2019. The work was done in connection with the BIOSTRATEG II programme and was aimed at a concrete industrial customer: a manufacturer of floor boards (Barlinek) that wanted to optimise how logs are cut into cants and planks. The authors report that the scanner handles logs of 250 to 500 mm diameter and up to 4000 mm long.

This article summarises what the authors built, how they calibrated and tested it, what accuracy they report, and what a sawmill planner can take from it. It is an independent summary prepared for the portal; the original paper should be consulted for any engineering decision. Where the portal adds its own reading, this is marked as interpretation.

Research methodology

Scanner hardware

The system is made of six identical measuring heads arranged around the log as it travels on a conveyor. Each head contains a laser line projector (a Z-LASER ZM18 laser ruler), an industrial camera from the Automation Technology C2 series, two optical mirrors and an aluminium profile frame. The optical axes of laser and camera are set at 30 degrees to each other, which is the geometric basis of laser triangulation: the camera sees the laser line shifted according to the height of the surface it falls on. The mirrors fold the optical path so that the head stays compact and is easier to align. Because sawmill yards and log decks are dusty, each head is protected by a pane of glass or similar material. The processing computer is a laptop-class machine with an Intel Core i7 3610QE processor at 2.30 GHz and 8 GB of RAM.

All six cameras are triggered at the same moment by an external synchronisation system, so that the six laser lines belong to the same cross-section of the log. Each camera is handled in its own software thread.

From images to points

The software extracts the laser line from each image either by taking the brightest pixel in each image column or by computing an intensity-weighted centre of gravity. A two-stage filter then cleans the data. The first stage removes isolated noise using a window size and a pixel-count threshold that the user can set. The second stage removes points that belong to the scanner casing or the conveyor, using a mask recorded from an empty conveyor with a user-defined tolerance. The remaining pixels are converted into 3D points by intersecting the camera ray for each pixel with the calibrated laser plane, and the points are moved into one global coordinate system.

The position of the log along the conveyor is obtained either from a laser rangefinder or from the trigger interval multiplied by conveyor speed.

Calibration

Calibration has two levels. In local calibration, each head is calibrated on its own. The camera intrinsic parameters (focal lengths, principal point and five distortion coefficients) are estimated using Zhang's flat-pattern method with circle-centre calibration targets. The laser plane is found by fitting a line with the RANSAC method to laser pixels seen on the pattern and then fitting a plane by least squares.

Global calibration ties the six heads together. The authors use a hexagonal artefact with a radius of 232 mm, with a different Charuco code on each of its six walls. Camera poses relative to the artefact are estimated with the EPnP algorithm, and a rotation correction aligns each camera's local x-axis with the conveyor direction. The authors point out that the conveyor partly hides the artefact from the lower heads, which weakens the pose estimates for those heads.

Cant optimisation

The paper also describes the software that turns the point cloud into a cutting proposal. Global and local cylinder-based filters remove noise, branches and bark remnants. A plane fitted to the cloud gives the starting orientation of the log. The software then places planks by checking for collisions between the cloud and cuboids that represent the products; colliding cuboids are shifted toward the log centre by a user-defined distance. The remaining planks are fitted into the side sections with a brute-force knapsack search that is limited by a time budget. The output includes laser marker positions and an image of the proposed cut edges for the operator. In the sample calculation, planks were 170 mm wide and 31 mm thick, with a 4 mm kerf.

Accuracy tests

Two reference tests were used:

• A ball-bar made of two bearing balls of 25 mm radius joined by a 400 mm steel bar, measured on a ZEISS ACCURA coordinate measuring machine (maximum permissible error 1.7 + L/333 micrometres). The scanner measured it in 10 scans for each of 3 orientations (parallel, perpendicular and oblique).
• A steel tube of 300 mm nominal diameter, compared with a reference scan from a Metris MCA 2 articulated arm with an MMCX80 laser scanner (uncertainty below 0.1 mm).

Key findings

The central result is that a six-head laser triangulation system can capture most of a log's surface in motion with an uncertainty of roughly two millimetres under the ball-bar test, and that profile processing is fast enough for on-line use. The larger maximum error of 8.02 mm on the tube is higher than the ball-bar result; the authors attribute it to noise, reflections and errors in the global calibration, and note that excluding two point clouds affected by conveyor occlusion brings the mean error to 1.86 mm. The conveyor also limits circumferential coverage of the sample log to 320 degrees rather than the full 360.

The cutting optimisation step needed between 10 and 35 seconds per log in the authors' tests, depending on the problem, whereas the geometry capture itself is a matter of milliseconds per profile. In practice, therefore, the optimisation, not the scanning, sets the cycle time of the whole system.

The authors tested the system at a Polish floor board manufacturer's sawmill. They conclude that it is suitable for use on a sawmill log line, while recommending better calibration and a different conveyor design for better results.

Practical applications for sawmills

What the paper itself supports

• Laser triangulation heads can be built to survive dusty sawmill conditions when protected by panes, and a multi-head arrangement can capture the outer log surface in one pass while the log moves.
• Scan data can be turned into a proposed set of cuts (cants and planks), with laser markers and an edge image for the operator, which makes the scan result directly usable at the log line.
• The paper's test product was a floor board component (170 x 31 mm planks), so the optimisation target there was a fixed product list.

Portal interpretation (not claims made by the authors)

• For a mill considering a log scanner, the paper suggests asking suppliers about calibration method, occlusion by the conveyor, and how the stated accuracy was verified. A ball-bar figure alone may not represent performance on a real, large, rough-surfaced log, as the tube result in this paper shows.
• A ten-to-thirty-second optimisation time may be acceptable on a slower breakdown line for large hardwood logs but could matter for high-speed softwood lines. Mills should check cycle times against their own throughput.
• The paper reports no yield or recovery figures, so the financial benefit of this specific system cannot be derived from it. Any business case needs a trial on the mill's own log mix.
• A conveyor that leaves the log's underside hidden will always limit coverage. When specifying a new installation, the transfer and conveyor design is part of the measurement system.

Limitations

The authors themselves list several weaknesses. The conveyor occludes the global calibration artefact and the log, which limits coverage to 320 degrees and degrades the calibration of the lower heads. The cameras are designed to extract laser profiles, not to take photographs, so images are noisy. The protective panes introduce reflections and distortion that affect calibration. Bundle adjustment of camera poses and laser planes was not yet applied and is planned for a future version.

Readers should also be aware of some points of inconsistency in the paper as we read it. The conclusions quote an accuracy of 2.4 mm, whereas the results section gives 2.16 mm, and the conclusions mention two sets of three scanners working in different wavelength regimes while the methods describe six identical heads. We could not resolve these differences from the text. In addition, some tables (local calibration errors and the product list) were not readable in the text version we accessed, so those values are not reported here. The tests were limited to a ball-bar, one steel tube and a sample log; the paper does not provide a statistical study across many logs, and it gives no data on bark condition, frozen or wet logs, or different species. Finally, the paper does not quantify the yield gain from the optimisation.

Conclusions

The paper shows a practical route to on-line 3D measurement of whole logs using six synchronised laser triangulation heads, a two-level calibration, and software that proposes cut layouts. Measured uncertainty was about 2 mm in the ball-bar test, with larger errors where the conveyor blocked the view, and profile processing took under 10 ms. The main engineering lessons are about calibration under occlusion, dust protection and conveyor design rather than about the laser technology itself. For sawmills, the work is a useful reference design and checklist, but it is not a proof of economic benefit, which still needs to be established mill by mill.

Original research

• Authors: Piotr Siekanski, Krzysztof Magda, Krzysztof Malowany, Jan Rutkiewicz, Adam Styk, Jakub Krzeslowski, Tomasz Kowaluk, Andrzej Zagorski
• Title: On-Line Laser Triangulation Scanner for Wood Logs Surface Geometry Measurement
• Journal: Sensors, 2019, volume 19, issue 5, article 1074
• DOI: 10.3390/s19051074
• Link: https://www.mdpi.com/1424-8220/19/5/1074
• Licence: Creative Commons Attribution (CC BY), open access

This article is an independent summary by the portal; the original paper should be consulted for full details.