I want to reconstruct the 3-D shape of an object based on a raw hand-drawn sketch. The sketch contains also the invisible lines of the object, and it is drawn on top of an already existing 3-D scene (of a city). I might be able to recover some 3-D points based on their coincidence with other objects from the city.
Reconstructed shapes do not need to be 100% precise, some inaccuracies are permitted, and they don’t contain curves and curved surfaces, only plain polygons.
I have found this topic:
generate 3d object from processing a 2d sketch on paper
but it seems that the problem there is not constrained at all, in contrast to my case. I also have read some papers, the newest one and most descriptive one is this one:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.118.2577&rep=rep1&type=pdf
The paper implements an algorithm based on constraints (e.g. parallel lines, parallel faces, symmetry, minimal distances, etc.). But it has no available implementation…
I would like to implement a fast-and-dirty parallelized algorithm, but I don’t seem to find any code to start from… I would appreciate any hint or link to an existing implementation. I know it is a hard problem, but I am sure it is not impossible to solve.
I found one solution to this problem, although not exactly the one I’m looking for:
Shape recognition and physics simulation
This approach recognizes shapes out of visible edges only. The user does not have to draw the invisible edges. I would like to use the invisible edges too.
Unfortunately, I couldn’t get access to the source code, even though they’ve given an e-mail for interest towards the source.