The goal is to have a module that can take a set of matching points from a set of n photographs and one or more reference distances to create an accurate 3D model. These matching points are manually done and not using feature detection. The objects are typically large buildings, both exterior and interior. There can be 50-100 images per set. We use a high res camera 7360x4912.
Previously we used the code from Chapter 4 of the Mastering OpenCV book.
See https://github.com/MasteringOpenCV. However the code does not produce accurate enough results. The existing code uses VS2012, opencv 2.4.6, and google test. I have several test files used to test accuracy.
I need a contractor to either produce a new module or correct the existing code. The contractor must be very familiar with Multiple view geometry and have created similar code.