Ask Your Question
1

solvePnP returning incorrect values

asked 2015-09-29 13:46:05 -0600

amb2189 gravatar image

updated 2015-10-06 18:14:21 -0600

I'm currently trying to implement an alternate method to webcam-based AR using an external tracking system. I have everything in my environment configured save for the extrinsic calibration. I decided to use cv::solvePnP() as it supposedly does pretty much exactly I want, but after two weeks I am pulling my hair out trying to get it to work. A diagram below shows my configuration. c1 is my camera, c2 is the optical tracker I'm using, M is the tracked marker attached to the camera, and ch is the checkerboard.

Diagram of my configuration

As it stands I pass in my image pixel coordinates acquired with cv::findChessboardCorners(). The world points are acquired with reference to the tracked marker M affixed to the camera c1 (The extrinsic is thus the transform from this marker's frame to the camera origin). I have tested this with data sets up to 50 points to mitigate the possibility of local minima, but for now I'm testing with only four 2D/3D point pairs. The resulting extrinsic I get from the rvec and tvec returned from cv::solvePnP() is massively off in terms of both translation and rotation relative to both a ground truth extrinsic I manually created as well as a basic visual analysis (The translation implies a 1100mm distance while the camera is at most 10mm away).

Initially I thought the issue was that I had some ambiguities in how my board pose was determined, but now I'm fairly certain that's not the case. The math seems pretty straightforward and after all my work on setting the system up, getting caught up on what is essentially a one-liner is a huge frustration. I'm honestly running out of options, so if anyone can help I would be hugely in your debt. My test code is posted below and is the same as my implementation minus some rendering calls. The ground truth extrinsic I have that works with my program is as follows (Basically a pure rotation around one axis and a small translation):

1     0     0     29
0   .77  -.64   32.06
0   .64   .77  -39.86
0     0     0     1

Thanks!

#include <opencv2\opencv.hpp>
#include <opencv2\highgui\highgui.hpp>

int main()
{
    int imageSize     = 4;
    int markupsSize   = 4;
    std::vector<cv::Point2d> imagePoints;
    std::vector<cv::Point3d> markupsPoints;
    double tempImage[3], tempMarkups[3]; // Temp variables or iterative data construction  

    cv::Mat CamMat = (cv::Mat_<double>(3,3) <<  (566.07469648019332), 0, 318.20416967732666, 
        0,  (565.68051204299513), -254.95231997403764, 0,  0, 1);  

    cv::Mat DistMat = (cv::Mat_<double>(5,1) <<  -1.1310542849120900e-001, 4.5797249991542077e-001,
    7.8356355644908070e-003, 3.7617039978623504e-003, -1.2734302146228518e+000);

    cv::Mat rvec = cv::Mat::zeros(3,1, cv::DataType<double>::type);
    cv::Mat tvec = cv::Mat::zeros(3,1,cv::DataType<double>::type);
    cv::Mat R;
    cv::Mat extrinsic = cv::Mat::eye(4, 4, CV_64F);

    // Escape if markup lists aren't equally sized
    if(imageSize != markupsSize)
    {
    //TODO: Replace with try, throw error code, and catch in qSlicerLocalizationModuleWidget
    return 0;
    }

    // Four ...
(more)
edit retag flag offensive close merge delete

Comments

Glad you solve your issue !

Eduardo gravatar imageEduardo ( 2015-10-07 07:26:58 -0600 )edit

1 answer

Sort by ยป oldest newest most voted
2

answered 2015-09-29 19:54:41 -0600

Eduardo gravatar image

updated 2015-09-30 04:37:15 -0600

(Note this answer was made before you add all your edit remarks.)

I think that a figure with all the transformation matrix and with all the frames could really help (you and/or us).


As I understand your problem, you have:

  • one camera, with an AR marker affixed on it, looking to a chessboard
  • for each chessboard corner, you have the 2D image coordinate and the 3D object coordinate with respect to the AR marker frame
  • you want to know the transformation between the camera frame and the marker frame

Setup1

Unless you know exactly the 3D coordinate of each chessboard corner with respect to the AR marker frame and unless the chessboard doesn't move, I don't see why it doesn't work.


If one of the two previous conditions is false, the setup should be in my opinion:

  • one camera, with an AR marker affixed on it, looking to a chessboard
  • another camera looking at the chessboard and the marker in the same time

Setup2


To obtain the transformation between the first camera and the marker: Equation

Anyway, a figure could help you if I did not understand your problem.

edit flag offensive delete link more

Comments

The second example you gave is the most accurate one to represent my system, although the second "camera" is actually a stereo-optical tracking system, so I can't acquire any pixel values from it (Which is fine because it returns transforms to the markers anyway!).

While I understand the equation you provided, I'm unclear on how to obtain c1Mch as Camera 1's pose is entirely unknown to the tracking system. Is that referring to the transform returned by cv::SolvePnP? If so, does that imply that I have the correct transform from PnP and I'm just manipulating it incorrectly while trying to obtain c1Mm?

Your diagram is really spot on, by the way. Can I add it to my question to help others visualize?

amb2189 gravatar imageamb2189 ( 2015-09-30 11:40:25 -0600 )edit
1

Yes you can obtain c1Mch with cv::SolvePnP. If you supply the list of 2D image coordinates and the list of 3D coordinates with respect to the chessboard frame, with the intrinsic parameters known, cv::SolvePnP will directly return the pose of the camera (c1Mch), no need to invert the matrix unless you want the inverse transformation of course.

You could verify it very easily: you already have the intrinsic camera parameters, cv::findChessboardCorners will return the list of corners and the 3D corner coordinates can be built easily (example). So you can check the ouput of cv::SolvePnP by printing the translation for example.

You can use the diagram if you want (with just a mention of my name please).

Eduardo gravatar imageEduardo ( 2015-09-30 12:54:50 -0600 )edit

Updated the question to include the diagram and added some descriptors (Also credited you at the bottom :) ).

The update contains some information that may complicate the process, so if you're willing to give it a go read the latest edit! I tried out your method as you described it and checked the math in Matlab after I obtained all the transforms but it didn't work. This is likely due to the new issue I've discovered though.

amb2189 gravatar imageamb2189 ( 2015-09-30 13:37:12 -0600 )edit

Question Tools

1 follower

Stats

Asked: 2015-09-29 13:46:05 -0600

Seen: 4,434 times

Last updated: Oct 06 '15