Setup a new viewport and use that one to render a top-down image of the level at startup, then delete the viewport. Then you can use the generated image as a background for the minimap and then just plot the player position and such.

The best way would really depend on what other data structures your RTS has lying around. For example, if your map file had a preview bitmap you might just load that and during collision detection (or some other phase where you need to cycle through all your actors) just stick pixels on the preview corresponding to the units and then display that. If you don't have something like that, and since you're using OpenGL you might render to an offscreen surface, once when the map is loaded, to generate the mini-map background and store that somewhere and use copies of that like if you had the preview bitmap.

Of course there are purists who would say that doing this during collision detection would be bad software engineering since you would be doing to logically unconnected things in the same code segment. YMMV.

Well, take that open gl camera and place it at the center of the level, then elevate it and point it downwards. Then "take a screenshot", store it as a surface, and then use that surface for the background of the minimap. I dont know opengl all that well but I know the routine you see. :)