Computing attitude(roll, pitch & yaw) from solvePnP()

edit

As @StevenPuttemans says, the out put of solvePnP() is not directly roll, yaw & pitch, but need to calculate using rotation matrix.

I also came through the same problem and found the function decomposeProjectionMatrix which gives euler angles as it's output.

Suppose we already get the rotation Mat from solvePnP()

solvePnP(..,...,rotCamerMatrix,transMatrix,false);
Rodrigues(rotCamerMatrix,rotCamerMatrix1);

Now using the function,

void getEulerAngles(Mat &rotCamerMatrix,Vec3d &eulerAngles){

    Mat cameraMatrix,rotMatrix,transVect,rotMatrixX,rotMatrixY,rotMatrixZ;
    double* _r = rotCamerMatrix.ptr<double>();
    double projMatrix[12] = {_r[0],_r[1],_r[2],0,
                          _r[3],_r[4],_r[5],0,
                          _r[6],_r[7],_r[8],0};

    decomposeProjectionMatrix( Mat(3,4,CV_64FC1,projMatrix),
                               cameraMatrix,
                               rotMatrix,
                               transVect,
                               rotMatrixX,
                               rotMatrixY,
                               rotMatrixZ,
                               eulerAngles);
}

Call it like,

   Vec3d eulerAngles;
   getEulerAngles(rotCamerMatrix1,eulerAngles);

where

yaw   = eulerAngles[1]; 
pitch = eulerAngles[0];
roll  = eulerAngles[2];

I think that what you doing wrong is considering that these angles are actually the exact values for roll, yaw & pitch. Looking at the documentation, the description states:

rvec – Output rotation vector (see Rodrigues() ) that, together with tvec , brings points from the model coordinate system to the camera coordinate system.

tvec – Output translation vector.

This means that for each point you have in your model, you can switch to the camera coördinate system. However, this is NOT the world coördinate system. I think you should still do a second operation to transfer between the camera coordinate system and the world coordinate system.

Maybe take a look at some topics that do exactly what you want, retrieving pitch yaw and roll.

1. http://www.morethantechnical.com/2010/03/19/quick-and-easy-head-pose-estimation-with-opencv-w-code/
2. http://planning.cs.uiuc.edu/node103.html