Wrong measure on Zaxis on Translation Matrix of SolvePnP

asked 2016-08-05 11:42:26 -0600

Diogo gravatar image

updated 2016-08-08 05:40:28 -0600

Hi, I am trying to get both rotation and translation matrix of a red square with webcam using SolvePnP. All the values are correct with my measures (with +-1mm of error) except the translation value of Z axis that gives me exacly half of the suppose value.


This is the code that give me the pose relatively to the center of the camera

void getPose(std::vector<std::vector<cv::Point> > contours)
    cv::Mat rvec;
    cv::Mat tvec;
    std::vector<cv::Point2f> image_points;
    std::vector<cv::Point3f> object_points;
    std::vector<cv::Point> contours_center;
    std::vector<cv::Point2f> contoursf;

    for( int i = 0; i < 4; i++ )

    std::cout << contours_center << std::endl;


    ///Model points in 3D
    object_points.push_back(cv::Point3f(25, 25,0));
    object_points.push_back(cv::Point3f(25, -25,0));
    object_points.push_back(cv::Point3f(-25, -25,0));
    object_points.push_back(cv::Point3f(-25, 25,0));

    ///Image points in the 2D image

    //Camera matrix obtained from camera calibration
    cv::Mat cam_matrix = cv::Mat(3,3,CV_32FC1,cv::Scalar::all(0));
    cam_matrix.at<float>(0,0) = 338.13;
    cam_matrix.at<float>(1,1) = 320.75;
    cam_matrix.at<float>(2,0) = 162.73;
    cam_matrix.at<float>(2,1) = 96.65;
    cam_matrix.at<float>(2,2) = 1;

    //Distortion matrix obtained from camera calibration
    cv::Mat dist_matrix = cv::Mat(1,5,CV_32FC1,cv::Scalar::all(0));
    dist_matrix.at<float>(0,0) = 0.0192;
    dist_matrix.at<float>(0,1) = -0.102;
    dist_matrix.at<float>(0,2) = -0.026;
    dist_matrix.at<float>(0,3) = -0.0026;
    dist_matrix.at<float>(0,4) = 0;

    ///Get rotation and translation matrix
    std::cout << "Rotaion: " << rvec << std::endl << std::endl;
    std::cout << "Translation: " << tvec << std::endl << std::endl << std::endl;

Result: link text Z distance in the translation axis is 158mm when it should be 320mm, the rest of values are correct

The green circle is the center of the image

edit retag flag offensive close merge delete



Can we see a sample picture and the relevant section of code?

Tetragramm gravatar imageTetragramm ( 2016-08-05 19:45:33 -0600 )edit

I edited the question with the code and a sample picture

Diogo gravatar imageDiogo ( 2016-08-08 03:20:31 -0600 )edit

if you mustuse imgur, please at least prefer http://i.imgur.com/xyz links, so folks here don't have to look at all the other shit.

berak gravatar imageberak ( 2016-08-08 03:26:29 -0600 )edit

Thank you Diogo. Let's check assumptions. You are working in mm, correct? So your square is 50x50 mm?

Next question. Why are you subtracting the contour points from the center point of the camera?

Last question for now. How certain are you of the camera matrix? If your image is 320x240, the principal Y seems farther off center than I'd expect.

Tetragramm gravatar imageTetragramm ( 2016-08-08 19:37:06 -0600 )edit

Yes, i am working in mm and the square is 50x50mm.

I am subtracting so i can get the contour position relatively to the center of the camera.

My image is 640x480. I am pretty sure the camera matrix is correct. I used a camera calibrator to find the matrix. Also i try already with other camera and the same error occurred.

Thanks for the help

Diogo gravatar imageDiogo ( 2016-08-09 03:53:30 -0600 )edit

I think i have found the problem, when i run the calibration program it uses a 320x240 image when the camera is in fact 640x480. Thanks for the help!

Diogo gravatar imageDiogo ( 2016-08-09 04:00:03 -0600 )edit