单张图片计算和下一张图片之间的单应矩阵--无需匹配点

使用场景-平面的映射（如地面，墙面）

单应矩阵的基础知识

链接: Basic concepts of the homography explained with code.

链接: Planar Homographies.

链接: 图像处理之理解Homography matrix(单应性矩阵).

OPENCV官方代码案例： https://github.com/opencv/opencv/tree/master/samples/cpp/tutorial_code/features2D/Homography

代码片

去博客设置页面，选择一款你喜欢的代码片高亮样式，下面展示同样高亮的 代码片.

// An highlighted block

void computeC2MC1(const Mat &R1, const Mat &tvec1, const Mat &R2, const Mat &tvec2,
                  Mat &R_1to2, Mat &tvec_1to2)
{
    //c2Mc1 = c2Mo * oMc1 = c2Mo * c1Mo.inv()
    R_1to2 = R2 * R1.t();
    tvec_1to2 = R2 * (-R1.t()*tvec1) + tvec2;
}
Mat computeHomography(const Mat &R_1to2, const Mat &tvec_1to2, const double d_inv, const Mat &normal)
{
    Mat homography = R_1to2 + d_inv * tvec_1to2*normal.t();
    return homography;
}
/*
* pose1 pose2 两个机器的POSE值（世界坐标系）
* extrinc_matrix 相机外参 4*4
* cameraMatrix_K 相机内参 3*3
* 输出：单应矩阵
*/
cv::Mat compute_homography_by_para(cv::Vec3f pose1,cv::Vec3f pose2 )
{

    double theta1 = pose1[2]   ;
    double theta2 =  pose2[2];
//    0.389087  Y:   A: -0.015  X: 0.05628  Y: -0.000273276  A: -0.007  LS: 0.381486  AS: 0
// X: 0.782331  Y: -0.024205  A: -0.052  LS: 0  AS: 0 0.799277  Y: -0.014131  A: -0.032
    cv::Mat M_v_o_1= (Mat_<double>(4,4) << cos(theta1),-sin(theta1),0.0,pose1[0],sin(theta1),cos(theta1), 0.0, pose1[1] , 0.0 , 0.0, 1, 0,  0.0, 0.0,0.0, 1);
    cv::Mat M_v_o_2= (Mat_<double>(4,4) << cos(theta2),-sin(theta2),0.0, pose2[0], sin(theta2),cos(theta2), 0.0, pose2[1], 0.0 , 0.0, 1, 0,  0.0, 0.0,0.0, 1);
    cv::Mat M_c1_o =   extrinc_matrix.inv() * M_v_o_1.inv();//extrinc_matrix * M_v_o_1   ;
    cv::Mat M_c2_o =  extrinc_matrix.inv() * M_v_o_2.inv();//extrinc_matrix * M_v_o_2 ;

    cv::Mat R_1to2, tvec_1to2;

    cv::Mat R1 = M_c1_o.rowRange(0,3).colRange(0,3).clone();
    cv::Mat tvec1 =  M_c1_o.rowRange(0,3).colRange(3,4).clone();

    cv::Mat R2 = M_c2_o.rowRange(0,3).colRange(0,3).clone();
    cv::Mat tvec2  = M_c2_o.rowRange(0,3).colRange(3,4).clone();

    computeC2MC1(R1, tvec1, R2, tvec2, R_1to2, tvec_1to2);


    cv::Mat normal = (Mat_<double>(3,1) << 0, 0, 1);
    cv::Mat normal1 = R1  * normal;
    cv::Mat origin(3, 1, CV_64F, Scalar(0));
    cv::Mat origin1 = R1 *origin + tvec1;
    double d_inv1 = 1.0 / normal1.dot(origin1);

    cv::Mat homography_euclidean = computeHomography(R_1to2, tvec_1to2  , d_inv1, normal1);
    cv::Mat homography = cameraMatrix_K * homography_euclidean * cameraMatrix_K.inv();
    homography /= homography.at<double>(2,2);
    return homography;
}
int main()
{

// 初始化机器移动的两个POSE值
        cv::Vec3f pose_vec, next_pose_vec;
        pose_vec[0]  = 0.389087 ;
        pose_vec[1] =  -0.024205 ;
        pose_vec[2]  =  0.05628   ;


        next_pose_vec[0] =  0.782331  ;
        next_pose_vec[1] =  -0.014205 ;
        next_pose_vec[2]  = -0.052 ;

        cv::Mat homography= compute_homography_by_para( pose_vec,next_pose_vec );

        Mat mask_output;
        warpPerspective(origin, mask_output, homography, Size(origin.cols,origin.rows));

        resize(mask_output, mask_output, Size(origin.cols/cutio, origin.rows/cutio));
        cv::Mat cmpare(mask_output, cv::Range(330 , image.rows ),cv::Range(0 , image.cols ));// cols
//        imwrite(convert.str(),image_next);
        cv::namedWindow( "correction", cv::WINDOW_AUTOSIZE );
        cv::imshow( "correction", cmpare      );
        cv::waitKey(0);
        return 0;
}