Is there any problem with passing image points to OpenCV's solvePnP function which do not lie inside the image itself? This includes points with negative coordinates.
If have read some tutorials/ answers where it was mentioned implicitely that the point should lie inside the frame but i can't think of any reason why this should be required.