RaymondAllen/asphalt-3d-laser-texture

Asphalt pavement 3D laser surface height maps with pendulum skid resistance (PTV)

Derivative of the Zenodo record 20606458, "Raw 3D laser scan point clouds and pendulum test value (PTV) and Traction Watcher One (TWO) friction measurements from 45 asphalt pavement test sections" by Matus Kovac, Matej Brna and Peter Pisca (University of Zilina, Faculty of Civil Engineering), https://doi.org/10.5281/zenodo.20606458, licensed CC-BY-4.0. This derivative is redistributed under the same licence with attribution to the original authors.

What is in it

• train/<id>.npz (720 files) and test/<id>.npz (180 files): one file per road-surface spot. Key z: a float16 array of shape (1520, 2520) holding surface heights in millimetres on a regular 0.05 mm grid covering the 126 x 76 mm contact area of the British Pendulum slider (2520 cells along the slider's slip direction, 1520 across it). Heights are relative to a least-squares plane fitted to the spot; unmeasured cells are NaN. float16 storage quantises heights to 0.002-0.004 mm at 2-8 mm.
• train_labels.csv: columns id, ptv for the 720 training spots. ptv is the Pendulum Test Value (EN 13036-4) measured on the wetted surface of the same spot, temperature-corrected, mean of the last five swings.
• patch_index.csv: id, split, sha256, shape, nan_fraction for every file.
• Training ids are <section>P<path><spot> (section 01-45, slip path 01/02, spot 01-10); test ids are opaque. The 180 test spots come from 9 sections with no spots in the training set; their labels are not published here.

How the height maps were produced from the raw record

For each spot the eight overlapping passes of the static laser profilometer (raw text point clouds, 0.015 x 0.050 mm sampling) were gridded at 0.05 mm, co-registered pairwise by cross-correlation, merged onto one 0.05 mm canvas (mean in overlaps), cropped to the 126 x 76 mm slider contact area, cleaned of isolated points and spikes and of cells steeper than the scanner's 60-degree limit, with small unmeasured gaps interpolated from neighbouring cells and the remaining gaps left as NaN, and levelled by subtracting a least-squares plane. The processing follows the chain the authors describe for this campaign. The code is published with the task that uses this dataset (data_preparation/).

Versions

• Third revision: the raised frame of the scanner at both ends of every pass (a plateau about 3 mm above the surface, several millimetres long) had survived into the first two revisions at one edge of most maps; every pass is now trimmed of its frame and sparse borders and referenced to its own plane before merging, the height offsets between passes are re-estimated as overlap medians, and the window is centred on the frame-free area (up to ~20 % of a map can now be unmeasured where the scans are short). The outermost passes carry a slow form error of a few millimetres relative to the fitted plane. All 900 height-map files change in this revision; train_labels.csv is unchanged and patch_index.csv carries the new digests.
• Second revision: the strip registration of two spots (05P0203, 45P0201) had failed in the first revision, leaving most of their maps unmeasured; both were re-registered and rebuilt. The other 898 height-map files were byte-identical to the first revision.
• First revision (2026-09-07): initial release.

Citation

Kovac, M., Brna, M., Pisca, P. (2026). Raw 3D laser scan point clouds and pendulum test value (PTV) and Traction Watcher One (TWO) friction measurements from 45 asphalt pavement test sections. Zenodo. https://doi.org/10.5281/zenodo.20606458