BEV image about cam2bev

[lreiher]

Note that the extrinsic camera matrix [R | t] describes how to transform points in world coordinates to camera coordinates and not the other way round. I found a short article explaining in more depth where the t = -RX comes from.

[loseyourself-gan]

Hi, Thank you! I got it.
I use the code to transform my own image. The result is like this :

I successfully transform the image, but they seem to in wrong direction . I need to rotate it to stitch the image. Dose it right ? Or I do something wrong . @lreiher

[lreiher]

Hard to say by just looking at those images.

If you need further assistance, could you please post the exact command that you are executing and also attach the camera config files that you are using?

[loseyourself-gan]

Actually, I use C++ to realize it. Here is my source code.the yml is same as your yml except for camera intrinsic matrix K.
The M is set to (1/50,0,-10,0,1/50,-10,0,0,0,0,0,1)
I use the camera extrinsic matrix and translation to realize it.
Thank you for your reply.

class camera{
public:
    camera(const char * _name){
        name = _name;
        std::string path = ymlpath + name;
        cv::FileStorage fs(path,cv::FileStorage::READ);
        if(!fs.isOpened()){
            std::cerr << "can not open the config yml!" << std::endl;
        }
        else{
            std::cout << "read config from the yml file" << std::endl;
        }
        cv::Mat K_,D_,R_,T_;
        fs["K"] >> K_;
        fs["D"] >> D_;
        fs["R"] >> R_;
        fs["T"] >> T_;
        cv::cv2eigen(K_,K);
        cv::cv2eigen(D_,D);
        cv::cv2eigen(R_,R0);
        cv::cv2eigen(T_,T);
        double roll,pitch,yaw;
        yaw = fs["yaw"];
        pitch = fs["pitch"];
        roll = fs["roll"];
        yaw = deg2rad(yaw);
        pitch = deg2rad(pitch);
        roll = deg2rad(roll);
        Eigen::Vector3d eulerAngle;
        eulerAngle << -roll,-pitch,-yaw;
        Eigen::AngleAxisd rollAngle(Eigen::AngleAxisd(eulerAngle(0),Eigen::Vector3d::UnitX()));
        Eigen::AngleAxisd pitchAngle(Eigen::AngleAxisd(eulerAngle(1),Eigen::Vector3d::UnitY()));
        Eigen::AngleAxisd yawAngle(Eigen::AngleAxisd(eulerAngle(2),Eigen::Vector3d::UnitZ()));
        Eigen::Matrix3d rotation_matrix;
        rotation_matrix=yawAngle*pitchAngle*rollAngle;
        cout << rotation_matrix.eulerAngles(0,1,2) << endl;

        R = rotation_matrix * R0;
        Eigen::Vector3d T1;
        T1 = -R * T;
        P.block<3,3>(0,0) = R;
        P.block<3,1>(0,3) = T1; 
        // P << R(0,0),R(0,1),T(0),
        //      R(1,0),R(1,1),T(1),
        //      R(2,0),R(2,1),T(2);
    };
    cv::Mat get_homography(Eigen::Matrix<double,4,3>&M,cv::Mat &image){
        Eigen::Matrix3d H,H_inv;
        cv::Mat h,dist;
        H = K * P * M;
        H_inv = H.inverse();
        cv::eigen2cv(H_inv,h);
        cv::warpPerspective(image,dist,h,image.size());
        return dist;
    }
    double deg2rad(double deg){
        return deg*M_PI/180;
    }
    cv::Mat undistort(cv::Mat & image){
        cv::Mat I0 = (cv::Mat_<double>(3,3)<< 1.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0,1.0);
        cv::Mat K_cv,D_cv,map1,map2,un_image;
        cv::eigen2cv(K,K_cv);
        cv::eigen2cv(D,D_cv);
        cv::Mat dist = K_cv.clone();
        cv::fisheye::initUndistortRectifyMap(K_cv,D_cv,I0,dist,cv::Size(1000,800),CV_16SC2,map1,map2);
        cout << K.inverse() << endl;
        cv::remap(image,un_image,map1,map2,cv::INTER_LINEAR);
        return un_image;
    }


private:
    Eigen::Matrix3d K;
    Eigen::Matrix3d R;
    Eigen::Matrix3d R0;
    Eigen::Vector4d D;
    Eigen::Vector3d T;
    Eigen::Matrix<double,3,4>P;
    static const std::string ymlpath;
    const char * name;

};

@lreiher

[lreiher]

Please understand that we cannot provide support for any work outside of our repository.

[loseyourself-gan]

Anyway , thank you for your reply. If I do it right, can I stitch the image directly ? And I need not to rotate it manually. @lreiher

[lreiher]

Yes, that should be possible with the extrinsic rotation matrix. Our code also doesn't rotate images manually.

[loseyourself-gan]

Thank you . I'll check it.