I spent several days to convince openGL to display the 3D model at the right position in the image given the camera and because when it comes to transformations usually 5 people have 5 different opinions I want to share the code here. For those looking for an explanation ... it is at my desk ... somewhere ... actually I totally forgot :)

I assume the camera coordinate system specified in the book multiple view geometry, i.e. x-axis to the right, y-axis down, z-axis facing away from the camera. If you use the MATLAB camera calibration toolbox, this is exactly the output you get.

Here is the code:

void updateViewport ()
{
	float u0 = // principal point x coordinate
	float v0 = // principal point y coordinate
	float fx = // focal x 
	float fy = // focal y
    
    // camera matrix should look like this
    // K = [ fx 0  u0
    //	 0  fy v0
    //	 0  0   1 ]
              
	float w =  // image width  in pixels 
	float h =  // image height in pixels
            
	glViewport ( 0,0,w,h);
 	glMatrixMode( GL_PROJECTION );
          
 	float proj[] = 	{	
	fx/(w/2),			0.0,			      0.0,					 0.0,
 	0.0,				fy/(h/2),		      0.0,					 0.0,
 	-(u0/(w/2)-1.0f),	(v0/(h/2)-1.0f),  (0.1f+1e6f)/(0.1f-1e6f),	 -1.0f,
 	0.0,				0.0,			  (2*0.1f*1e6f)/(0.1f-1e6f), 0.0 
 	};
    
 	glLoadIdentity (); 
              
	if ( fx != 0 && fy != 0) 
 		glLoadMatrixf ( proj );
    
	glMatrixMode ( GL_MODELVIEW );
	glLoadIdentity ();
}