Why SVD is required in estimation of homography matrix using RANSAC method? & what is the input for SVD?

6 次查看（过去 30 天）

swathi 2018-3-11

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/387603-why-svd-is-required-in-estimation-of-homography-matrix-using-ransac-method-what-is-the-input-for

评论： swathi 2018-3-19

Hello, I have tried a code given for imagestitching. But I am getting error in the following function: in P matrix calculation.

function [P,S,V,q,A] = estimateTransform(img1_points,img2_points)

    %%Initialization
    %format long
    ncorr = length(img1_points);
    %%Setting the P matrix 10x9
    P = [img1_points(1:ncorr,:),ones(ncorr,1),zeros(ncorr,3),...
        -1*img2_points(1:ncorr,1).*img1_points(1:ncorr,:),...
        -1*img2_points(1:ncorr,1);...
        zeros(ncorr,3),-1*img1_points(1:ncorr,:),-1*ones(ncorr,1),...
        img2_points(1:ncorr,2).*img1_points(1:ncorr,:),...
        img2_points(1:ncorr,2)];
    %%Calculating r 8x1
    %r = img2_points(:);
    %%SVD Decomposition
    [~,S,V] = svd(P);
    %%Extracting Diagonal elements of S
    sigmas = diag(S);
    %%Detecting minimum diagonal element
    if length(sigmas) >= 9 %trivial solution
        minSigma = min(sigmas);
        [~,minSigmaCol] = find(S==minSigma);
        %%Calculating q
        q = double(vpa(V(:,minSigmaCol)));
    elseif length(sigmas)<9 %non-trivial solution
        %%Calculating q
        q = double(vpa(V(:,9)));
    end
    %%Calculating A
    A = reshape(q,[3,3])';
end