Estimate 2D transformation between two sets of points using RANSAC

Question

As I know, OpenCV uses RANSAC in order to solve the problem of findHomography and it returns some useful parameters like the homograph_mask.

However, if I want to estimate just 2D transformation which means an Affine Matrix, is there a way to use the same methodology of findHomography which uses RANSAC and return that mask ?

related: http://answers.opencv.org/question/6747/is-there-a-way-to-compute-a-ransac-based-affine-transformation/ — Miki, Oct 15 '15 at 09:36
[Here](http://stackoverflow.com/questions/14833049/how-to-estimate-2d-similarity-transformation-linear-conformal-nonreflective-si) it is mentioned to use [estimateRigidTransform](https://github.com/Itseez/opencv/blob/ddf82d0b154873510802ef75c53e628cd7b2cb13/modules/video/src/lkpyramid.cpp) that seems to use RANSAC internally, — Miki, Oct 15 '15 at 09:41
if you know how to compute a 2D transformation from 3 point pairs, you can easily code your own simple RANSAC. Afaik, estimateRigidTransform uses ransac, but unfortunately you can't parametrize it... — Micka, Oct 15 '15 at 15:27

ZettaCircl · Accepted Answer · 2019-05-10T12:42:56.230

You can directly use estimateAffinePartial2D : https://docs.opencv.org/4.0.0/d9/d0c/group__calib3d.html#gad767faff73e9cbd8b9d92b955b50062d

cv::Mat cv::estimateAffinePartial2D (   
    InputArray  from,
    InputArray  to,
    OutputArray     inliers = noArray(),
    int     method = RANSAC,
    double  ransacReprojThreshold = 3,
    size_t  maxIters = 2000,
    double  confidence = 0.99,
    size_t  refineIters = 10 
)

for example :

        src_pts = np.float32([pic1.key_points[m.queryIdx].pt for m in matches]).reshape(-1, 1, 2)
        dst_pts = np.float32([pic2.key_points[m.trainIdx].pt for m in matches]).reshape(-1, 1, 2)

        # Find the transformation between points, standard RANSAC
        transformation_matrix, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)

        # Compute a rigid transformation (without depth, only scale + rotation + translation) and RANSAC
        transformation_rigid_matrix, rigid_mask = cv2.estimateAffinePartial2D(src_pts, dst_pts)

score 7 · Answer 2 · answered Aug 16 '16 at 19:06

estimateRigidTransform does use RANSAC internally, though the parameters are fixed at the moment - see the code here - https://github.com/opencv/opencv/blob/master/modules/video/src/lkpyramid.cpp

cv::Mat cv::estimateRigidTransform( InputArray src1, InputArray src2, bool fullAffine )
{
    const int RANSAC_MAX_ITERS = 500;
    const int RANSAC_SIZE0 = 3;
    const double RANSAC_GOOD_RATIO = 0.5;

    // ...

    // RANSAC stuff:
    // 1. find the consensus
    for( k = 0; k < RANSAC_MAX_ITERS; k++ )
    {
        int idx[RANSAC_SIZE0];
        Point2f a[RANSAC_SIZE0];
        Point2f b[RANSAC_SIZE0];

        // choose random 3 non-complanar points from A & B
        for( i = 0; i < RANSAC_SIZE0; i++ )
        {
            for( k1 = 0; k1 < RANSAC_MAX_ITERS; k1++ )
            {

The code had changed since your copy-paste. It directly calls estimateAffinePartial2D now. — ZettaCircl, May 08 '19 at 09:59

Estimate 2D transformation between two sets of points using RANSAC

2 Answers2