在OpenCV中对齐两个图像 [英] Align two images in OpenCV

问题描述

我有两张图片(见下文).这些图像代表一对电缆的轮廓，并使用基于激光的3D三角测量法捕获.第一个图像是用左摄像头捕获的，而第二个图像是用右摄像头捕获的.可以看出，这些图像是部分重叠的.第一图像上的左部分部分地对应于第二图像上的左部分.正确的部分也是如此.我想将这两个图像合并为一个图像，以使相应的部分重叠.

I have two images (see below). These images represent the contours of a pair of cables and were captured using laser based 3D triangulation. The first image is captured with the left camera, while the second one with the right camera. As can be seen, these images are partially overlapping. The left part on the first image partly corresponds to the left part on the second image. The same holds for the right part. I want to merge these two images into one image so that the corresponding parts overlap.

在这些图像旁边，我还可以使用以下信息:

Next to these images, I also have the following information at my disposal:

• 左右摄像机的3x3单应矩阵 H
• 左右摄像机的固有摄像机参数 K
• 左右摄像机的失真系数 D (其中9个)
• 左右摄像机的偏移 O

• A 3x3 homography matrix H of left and right camera
• Intrinsic camera parameters K of left and right camera
• Distortion coefficients D (9 of them) of left and right camera
• Offset O of left and right camera

此数据在下面指定.

在Halcon中，我尝试使用镶嵌法做到这一点:

In Halcon, I have tried to do this using mosaicking:

• 使用Harris提取两幅图像中的特征点
• 使用Ransac计算从一幅图像到另一幅图像的投影变换矩阵.
• 应用找到的投影变换矩阵.

但是这种方法并不成功.我正在OpenCV或Halcon 或中寻找一种类似的方法(也在OpenCV或Halcon中)，该方法利用了我掌握的校准数据，例如单应性矩阵和相机矩阵.

This approach was however not successful. I am looking for a similar approach in OpenCV or Halcon or an approach (also in OpenCV or Halcon) that makes use of the calibration data I have at my disposal, such as the homography matrix and camera matrix.

如果可能的话，请提供足够的解释，因为我只是从Machine Vision开始.

Please provide ample explanations, if possible, since I am only starting out with Machine Vision.

Hl := [0.00175186,   4.73083e-05, -0.00108921,
       0.000780817, -0.00145615,   0.00118631,
       0.0534139,   -0.030823,     1.0        ]
Kl := [4578.21,   -5.05144, 759.766,
       0.0,     4576.87,    568.223,
       0.0,        0.0,       1.0   ]
Dl := [-0.12573, 0.0533453, -0.575361, -0.0130272, 0.00348033, 0.00852617, -0.0271142, 0.0176706, -0.00575124]
Ol := [0.0, 150.0]

Hr := [0.00173883, -2.94597e-05, 0.00109873,
      -0.00077676, -0.0014687,   0.00121393,
      -0.0653829,  -0.0443924,   1.0        ]
Kr := [4591.96,  -4.55317, 1284.74,
       0.0,    4591.19,     534.317,
       0.0,       0.0,        1.0   ]
Dr := [-0.110751, -0.349716, 3.86535, 0.017393, -0.00364957, -0.00633656, 0.0338833, -0.0212222, 0.00543694]
Or := [0.0, 100.0]

推荐答案

模板匹配可以解决问题.我玩了一点，希望对您有用(下面的代码):

Template matching would do the trick here. I played with it a little, hope you find it usuful (Code below):

MAX_DISPARITY = 100;
imgL=double(imread('https://i.stack.imgur.com/y5tOJ.png'));
imgR=double(imread('https://i.stack.imgur.com/L1EQy.png'));
imgRfused = imgR;
minmax = @(v) [min(v) max(v)];
[imgLbw,n]=bwlabel(imgL);
nBlobs=2;
a=arrayfun(@(i) sum(imgLbw(:)==i),1:n);
[~,indx]=sort(a,'descend');
imgLbwC=bsxfun(@eq,imgLbw,permute(indx(1:nBlobs),[3 1 2]));
imgLbwC =bsxfun(@times,imgLbwC,2.^permute(0:nBlobs-1,[3 1 2]));
imgLbwC  = sum(imgLbwC ,3);

src = zeros(nBlobs,4);
dst = zeros(nBlobs,4);

for i=1:nBlobs
    [y,x]=find(imgLbwC==i);
    mmx = minmax(x);
    mmy = minmax(y);
    ker = imgL(mmy(1):mmy(2),mmx(1):mmx(2));

    [yg,xg]=ndgrid(mmy(1):mmy(2),mmx(1):mmx(2));
    src(i,:)=[mmx(1) mmy(1) fliplr(size(ker))];


    imgR_ = imgR(:,mmx(1)-MAX_DISPARITY:mmx(2)+MAX_DISPARITY);
    c=conv2(imgR_ ,rot90(double(ker),2),'valid')./sqrt(conv2(imgR_.^2,ones(size(ker)),'valid'));
    [yy,xx]=find(c==max(c(:)),1);
    dst(i,:)=[src(i,1:2)+[xx yy-mmy(1)]+[-MAX_DISPARITY,0] fliplr(size(ker))];

    imgRfused(dst(i,2):dst(i,2)+dst(i,4),dst(i,1):dst(i,1)+dst(i,3)) = max(imgRfused(dst(i,2):dst(i,2)+dst(i,4),dst(i,1):dst(i,1)+dst(i,3)),imgL(src(i,2):src(i,2)+src(i,4),src(i,1):src(i,1)+src(i,3)));
end
imagesc(imgRfused);
axis image
colormap gray

这篇关于在OpenCV中对齐两个图像的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！