Just for the record.
The current stitching utility in open cv does not consider translation of the camera and only assumes that camera is rotated around its axis. So, basically it tries to create and project images on cylindrical or spherical canvas.
But in my case, I needed to consider the translation motion while predicting the camera transformation and this is not possible with the existing stitching utility of opencv.
All these observations are made based on the code walk-through of opencv code and through trials.
But, you are welcome to correct this information or to add more information so that this can be a useful future reference.
Regards