Pose estimation from arbitrary images

edit

I am using an arbitrary image and using the template matching / homography example provided by opencv to find the known image in the another image and find bounding boxes. I would further like to estimate the pose of the camera using the solvePnP or similar method. However I am confused as to how I can get the objectPoints in 3-D space and their respective 2D translations in camera image.

How can I use cv::solvePnP to get the pose ? I have the calibrated the camera and have the distortion coefficients, and the camera matrix. How can I generate the objects points (3D model) from the following original image and get the same points in 2D in the detected image?