» Validation

We scanned a lot of objects with our Kinect and we want to evaluate the quality of our scans. We decided us to validate the scan results. To be able to classify the scanned objects, we assign them to specific properties.

object size
(mm)
surface structure conclusion
smooth rough glossy matt coarse fine
croktoy 270x267x150 yes no no yes yes no perfect object to scan
topscorer 70x70x150 yes no yes no no yes impossible to scan because of the reflective surface
duck 70x50x40 yes no no yes yes no measuring tolerance are noticed (merging the point clouds) because the object is to small
pig 200x150x150 yes no no yes yes no perfect object to scan
only ‘spring legs’ get lost
phone 220x170x100 yes no yes no yes no perfect object to scan
robotarm 200x100x200 yes no no yes no yes because of the fine structure, many details get lost
bunny 300x300x150 no yes no yes yes no perfect object to scan
Overhead Projektor 500x500x600 yes no no yes yes no in the height the pointclouds are drifting apart

After a few scans we noticed that the size of the scanned objects is an important factor. If the object is to small, the merging of the pointclouds is not perfect. But if it is to big, it is problematic too, because farther away from the origin the point clouds drifting more apart.

On the one hand the surface texture (smooth/rough) is almost meaningless, but on the other hand the reflectivity is very important. An object with a high reflection send us lesser depth information.

A course structure is a good precondition for a good result in contrast to a fine structure. A lot of details get lost by scanning objects with a fine structure.

All in all an object with the size and a broadly similar shape of a football is a perfect item to scan.