When the view camera is an orthogonal one (like it is for the top, front, left etc. cameras) then it is relatively straight forward.
When the camera is a perspective one, then it's a bit trickier as the picking volume is really a frustrum (truncated pyramid), but the general idea is the same.
If picking is being done along a line, then the math is simpler (or rather there's just less of it) for determining whether any particular polygon is intersects the pick ray (the need to take into account the camera viewing transform is still present however).