This post treats the reconstruction of the matching points and the results of this process.

Triangulation

Once the correspondent points are known, it’s possible to reconstruct the 3D model. The fast way is using the OpenCV function triangulatePoints over the points with a low disparity value. Disparity is the difference on coordinates between the left point and the correspondent point.

Another possible method is using the equations of the points, where f is the focal distance, B is the baseline or distance between cameras, d is the disparity and u and v are the coordinates of the point referred to the center of the image:

Z = f * B / d

X = u * Z / f

Y = v * Z / f

The results of this method are shown in the following sequence, where all the points are considered and it is reconstructed in row order:

Final result

The final result is the next one, which can be overlapped with a real image of the 3D scene, with a decent amount of correspondence, specially over the right blocks.

Thanks for reading!