How To Estimate 4 Parameters Camera Model?

Recently I am doing a project in image stitching. I want to carry out my project based on the paper "Automatic Panoramic Image Stitching using Invariant Features" by David G. Lowe and Matthew Brown. In their works, they used a 4 parameters camera model including one fixed focal length for all cameras, 3 parameters concerning rotation. They first estimate the parameters by direct linear transformation (DLT), and then do sparse bundle adjustment later. It seems that, we can solve the 3 rotation parameters given the focal information by DLT, but How do they solve the 4 parameters by DLT? Detailed mathematical formation will be much helpful. Do appreciate your help!

Thanks