Building Computer Vision Projects with OpenCV 4 and C++

Random sample consensus (RANSAC)

In the preceding image, we illustrate the fact that not all points conform to the affine constraint, and most of the matched pairs are discarded as incorrect. Therefore, in most cases we employ a voting-based estimation method, such as random sample consensus (RANSAC), where a group of points is chosen at random to solve for a hypothesis of M directly (via a homogeneous linear system) and then a voting is cast between all points to support or reject this hypothesis.

The following is a pseudo-algorithm for RANSAC:

1. Find matches between points in image i and image j.
2. Initialize the hypothesis for the transform between image i and j, with minimal support.
3. While not converged:
   1. Pick a small random set of point-pairs. ...