Problems reconstructing simple object

[kevin284]

Hi, I tried to reconstruct the 3D model of a simple lego minifigure on a non-reflective white background (A4 sheet) but doesn't matter all the work I did to mask out the white region (I even tried masking a few of them or not masking at all), I couldn't get any decent 3D mesh out of it.

The photo set is here: http://imgur.com/a/SdKI5

While this is the result I got: http://imgur.com/cgHCmfA

I was considering buying a license to reconstruct similar size objects but this seems to be a bummer. I followed the instructions on how to reconstruct the object on the english tutorial.

Can anyone shed some light on why this is happening please?

[JarkkoL]

Hi.

Take lot more pictures, perhaps something like one every 10-20 degrees, images must have enough overlap for this to work.

The surface you are trying to scan has very low surface variation in color and has strong subsurface scattering.
It's quite hard for software find any good tracking points in the image as highlights move on surface, shadow doesn't really move in background and object moves etc.

If you only want good surface and not color you can dust the object with powder and then scan, gives a surface that can be tracked properly. (Paint object later in software of your choice.)


Try to use background image to use as mask.

Add small non obstructing different colored objects around the target on rotating disk.
Anything goes as long as long as they are something the software can easily track and are not in your background masking image, ie small rocks glued to the A4.
This gives you better camera locations in scene, quite helpful trick in outdoors as well. (Usually throw couple of different colored balls around target.)

I haven't really done lot with object rotation methods, taking shots from different angles is easier to do test shots at the beginning. (Everything moves properly in the image as they should.)

[frank.stremke]

as said before you need something for photoscan to hold on to
use milimetric paper with soem random sribls on it or a non refective newspaper page instead of the a4
the newspaper will be modeled as well but you can cut it of
frank

[Heinrich]

I'd do as the previous poster suggested: use a highly structured background, so that Photoscan has a chance to align the images even if the minifig gives too few tie points. Also, you can place the fig on a small perspex cube or a smooth-sided box of styrofoam or so - makes cutting off the unwanted stuff easier.

Do take a LOT of photos! The worse the desired surface is for feature detection, the more images will you need to get a good model.