3D scan quality
This is a very first test of a comparison, please see the discussion below. Three different setups have been used to create the shown 3d models
- first column: OpenScan Pi with Pi Camera (8 megapixel) + ringlight+ polarizer
- second column: OpenScan Pi with external camera Fuji S100FS (11 megapixel) + ringlight+ polarizer
- third column: Thunk 3D Cooper structured light scanner
It is not a simple task to judge the quality of a scan. It is even more complicated to compare different scanning systems
There are various parameters influencing the result: resolution, accuracy, statistical spread, scan volume, captured features (e.g. holes, undercuts), experience of the operator and the object itself.
It would be great to have a neutral benchmark showing the advantages and disadvantages of different systems. I would be very happy to do the presentation and evaluation of the results. Anyway the discussion should be neutral and open to all. Therefore I need the help of some other scan-ethusiasts, who are willing to share some of their time creating a scan
As soon as there is a handful of scans available, I will create a separat website. Feel free to contact me at firstname.lastname@example.org.
In order to achieve comparability, it is absolutely necessary to have challenging scan objects. In my opinion the following points are crucial:
- no copyright or trademark
- challenging features (holes, undercuts, many details on different scales)
- various sizes
These could be some choices:
- 1.5-2cm - small coin
- 6cm 3d printed Benchy (see discussion below)
- 20cm+ ???
(as I am still looking for larger benchmark-candidates, feel free to share your ideas :)
The shown "Benchy"-Boat has been printed on a FDM printer with 0.2mm layer height. I've used the original CAD (see Thingiverse) as a reference. Surely most deviation from the CAD are caused by 3d printing itself. Nevertheless there are several differences visible in the scan results, which are clearly caused by the scan procedure itself.
3D printing created several features, which are unique in each and every print, there are blobs, layerlines, stringing ... and many more small and tiny features on the surface of the object. Of course, every printer would create a slightly different object, but nevertheless the question is, whether the scanner is able to pick up those impurities...
The first two models are based on 288 photos, 36 per rotation in 8 layers. Reconstruction has been done in Reality Capture. The difference in quality is caused by the overall image quality and different resolution of the cameras.
Photogrammetrie offers the great advantage, that the angle between two viewpoints can be freely chosen, which allows to capture small holes and undercuts better. Structured light and laser line scanning have a fixed angle between lightsource and camera and thus are limited to a certain degree. This is clearly visible on the inner areas of the model, which is very challenging to capture properly.
Ignoring the red (inaccurate) areas of the scans, the overall accuracy/scattering lies within 0.1 and 0.2mm. This is not the generell accuracy of the scanning system as the deviation mostly comes from the 3d printing process.
This discussion definitely needs to be continued. It will be really interesting to see the comparison to high-end and very low-end scanners. I hope that some people might share some of their time for this project. I would even send you a printed Benchy in exchange for the scan result. So feel free to contact me any time :)