3D Scanning Technologies

3D scanning is the process of creating a virtual model of a real object. There are many techniques, mainly split based on whether they have contact with the object being scanned or not. I only had access to non-contact technologies, and will be comparing two of them.


HP Z 3D

HP Z 3D is an overhead scanner attached to a computer, and performs the scan by having the user rotate the object. The computer creates the scan proceduraly as it sees more sides. It starts by placing the object on a plain white surface and then asking the user to rotate it slowly. Unfortunately the camera often loses tracking of the object and requires you to restart. A fairly simple scan (a Rubik's cube) took about 25 tries and close to 40 minutes. The final quality can be seen here:

The software also offers to create texture maps of the object and apply them in the end. This process never quite worked out for me, and I got blurry color mixtures on all sides.

Photogrammetry

Photogrammetry is the process of automatically creating a model based on pictures. It works by matching points between multiple pictures to figure out the structure.

In my experiment, I took pictures from 3 different elevations (0 , 35, and 60 degree angles). At each elevation, I was advised to take a picture every 10 degrees, meaning I took 36 photos per elevation. I also took photos without the cube on them to help the matching algorithm figure out the background information.

If everything works well, the software manages to align the cameras and create a sparse point cloud in a few minutes. If some images don't get aligned, the program asks the user to create matching points between pictures. In my case one elevation got aligned perfectly, and I did manual alignment of the other two. Manual alignment can be frustrating and time consuming - it took me well over 4 hours of realigning the pictures to get a satisfying result.

The dense point cloud was used to create a mesh of the cube, and the quality was impressive. If you zoom in, you could clearly see the separations between the little cubes. Along with it, the texture maps extracted had clear, correctly mapped colors.

Overall, I am pretty happy with the result. Despite the process being automatic, photogrammetry still required a lot of tedious work, both with taking the photos and with manually aligning them.