特征提取与匹配基础矩阵单应矩阵极限约束

Posted yanfang20180701

tags:

篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了特征提取与匹配基础矩阵单应矩阵极限约束相关的知识,希望对你有一定的参考价值。

 

Ubuntu16.04 + opencv2.4.9

一、特征提取与匹配

(以ORB特征为例)

features.cpp

 1 #include<iostream>
 2 #include<opencv2/core/core.hpp>
 3 #include<opencv2/highgui/highgui.hpp>
 4 #include<opencv2/features2d/features2d.hpp>
 5 #include<opencv2/calib3d/calib3d.hpp>
 6 using namespace cv;
 7 int main(int argc, char* argv[]){
 8 
 9 if (argc!=3){
10     std::cout << "usage: feature_extraction imge1 image2" << std::endl;
11     return 1;
12 }
13 //read image
14 Mat image1 =imread(argv[1],CV_LOAD_IMAGE_COLOR);
15 Mat image2 = imread(argv[2],CV_LOAD_IMAGE_COLOR);
16 Mat describtor1, describtor2;
17 vector<KeyPoint> keypoint1,keypoint2;  
18 //step2  orb 
19 ORB orb;
20 orb(image1,Mat(),keypoint1,describtor1);
21 orb(image2,Mat(),keypoint2,describtor2);
22 Mat outimage1;
23 drawKeypoints(image1,keypoint1,outimage1,Scalar::all(-1),DrawMatchesFlags::DEFAULT);
24 namedWindow("orb特征点",0);
25 cvResizeWindow("orb特征点", 640,480);
26 imshow("orb特征点",outimage1);
27 //step3 match
28 vector<DMatch> matches;
29 BFMatcher matcher(NORM_HAMMING);
30 matcher.match(describtor1,describtor2,matches);
31 //step4 find min dist and max dist of all matches
32 double mindist=1000,maxdist=0;
33 for(int i=0; i<describtor1.rows;i++){
34     double dist =matches[i].distance;
35     if(dist<mindist) mindist = dist;
36     if(dist>maxdist) maxdist = dist;
37 
38 }
39 std::cout <<"Max="<<maxdist <<std::endl;
40 std::cout<<"Min=" <<mindist <<std::endl;
41 //5 
42 std::vector<DMatch> good_matches;
43 for(int i=0; i<describtor1.rows; i++){
44    if(matches[i].distance <= max(2*mindist,30.0))
45    good_matches.push_back(matches[i]);
46 }
47 //step6 plot results
48 Mat img_match,img_goodmatch;
49 drawMatches(image1,keypoint1,image2,keypoint2,matches,img_match);
50 drawMatches(image1,keypoint1,image2,keypoint2,good_matches,img_goodmatch);
51 namedWindow("所有匹配点",0);
52 namedWindow("优化后匹配点",0);
53 cvResizeWindow("所有匹配点", 1021,376);
54 cvResizeWindow("优化后匹配点", 1021,376);
55 imshow("所有匹配点",img_match);
56 imwrite("img_match.png",img_match);
57 imshow("优化后匹配点",img_goodmatch);
58 imwrite("img_goodmatch.png",img_goodmatch);
59 waitKey(0);
60 return 061 }

结果:

Max=94
Min=21

orb特征点:

 技术图片

所有匹配点:

技术图片

 

 优化后匹配点:

 技术图片

 

 二、基础矩阵、单应矩阵

(调用opencv函数) 

 1 vector<Point2f> points1;
 2 vector<Point2f> points2;
 3 for (int i=0; i<matches.size();i++){
 4     points1.push_back(keypoint1[matches[i].queryIdx].pt);
 5     points2.push_back(keypoint2[matches[i].queryIdx].pt);
 6 }
 7 Mat fundamental_matrix;
 8 fundamental_matrix = findFundamentalMat (points1,points2,CV_FM_8POINT);
 9 std::cout <<"Fundamental_matrix="<< fundamental_matrix<<std::endl;
10 
11 Mat homography_matrix ;
12 homography_matrix= findHomography (points1,points2,RANSAC);
13 std::cout << "homography_matrix="<<homography_matrix <<std::endl;

结果:

Fundamental_matrix=[-4.075224349747993e-09, 2.477107638875577e-06, -0.0003483120020484568;
  -5.388675340517034e-08, 3.801248867773318e-05, -0.005352597475279493;
  2.16130923207796e-05, -0.007172831622169928, 1]
Homography_matrix=[-0.2147533899810728, -1.588025524342637, 437.941430304518;
  -0.07845693563076726, -0.580883857003762, 160.1035190035557;
  -0.0004891439919827992, -0.003633821403150171, 1]

三、极线约束

 1 //极线约束
 2 std::vector<Vec<float,3> > epilines1,epilines2;
 3 computeCorrespondEpilines(points1,1,fundamental_matrix,epilines1);
 4 computeCorrespondEpilines(points2,2,fundamental_matrix,epilines2);
 5 RNG rng;
 6 for(uint i=0;i<10;i++){
 7     Scalar color = Scalar(rng(256),rng(256),rng(256));
 8     circle(image2,points2[i],6,color,6);
 9     line(image2,Point(0,-epilines1[i][2]/epilines1[i][1]), Point(image2.cols,-(epilines1[i][2]+epilines1[i][0]*image2.cols)/epilines1[i][1]),color);
10     circle(image1,points1[i],6,color,6);
11     line(image1,Point(0,-epilines2[i][2]/epilines2[i][1]), Point(image1.cols,-(epilines2[i][2]+epilines2[i][0]*image1.cols)/epilines2[i][1]),color);
12 }
13 namedWindow("epilines1",0);
14 resizeWindow("epilines1",640,480);
15 imshow("epilines1",image1);
16 imwrite("epilines1.png",image1);
17 namedWindow("epilines2",0);
18 resizeWindow("epilines2",640,480);
19 imshow("epilines2",image1);
20 imwrite("epilines2.png",image1);
21 waitKey(0);

结果:

技术图片

技术图片

 

以上是关于特征提取与匹配基础矩阵单应矩阵极限约束的主要内容,如果未能解决你的问题,请参考以下文章

对极几何的理解和原理推导

opencv cv2.findHomography sift、surf特征匹配 单应性矩阵理解

SLAM面试题,单应矩阵基础矩阵F本质矩阵E之间的区别与联系

OpenCV单应性矩阵发现参数估算方法详解

国科大人工智能学院《计算机视觉》课 —底层视觉—特征点提取描述与匹配

国科大人工智能学院《计算机视觉》课 —底层视觉—特征点提取描述与匹配