Sandpaper Wind Turbine Blade Benchmark Dataset
The proposed dataset aims to provide a easy way for benchmarking Structure from Motion methods for capturing micro and macro structure details. The tested scenario is the use of SfM for capturing the surface roughness and shape of a wind turbine blade, but the same testing scenarios can be used for other use cases, where both the overall shape and the micro roughness of the surface are necessary, like inspection of tunnels, facades, sewers, etc. The dataset consists of two parts - a height artefact scenario, aimed to test the capturing capacity for macro resolution in SfM software. A custom height artefact is produced and used for this scenario, consisting of a number of heights from 10 mm to 0.625 mm. A number of images are present in the dataset of the artefact - both with drawn and projected noise for easier reconstruction. In addition the ground truth for the measurements of the artefact is present using a HIROX microscope. Finally, the STL used for 3D printing the artefact is also present, for easier testing. The second scenario is directed at detecting micro surface roughness. It is shown that sandpaper roughness can be used for approximating surface roughness, as it is easily quantifiable and has a known ISO standard. The used sandpaper has a ISO 6344 standard, given here - https://www.iso.org/standard/12643.html . A number of sandpaper grits are used - P40, P60, P80, P100, P120, P180 and P240. Each of the sandpaper patches is mounted on a wind turbine blade replica CnCed from Styrofoam, with the used STL also present in the dataset. Each sandpaper grit size is imaged with 57 images in a semi-circle with 3 horizontal bands of 19 images with different height. All images are captured using Canon 6D DSLR camera. All images contain EXIF data with used camera parameters. In addition, each of the sandpaper grits contains ground truth data captured with the same Hirox RH-2000 microscope. The ground truth is in absolute scale and contains additional data directly in the CSV files.