Sorry for the initial lack of detalis, here I go:
Two cameras are placed together and are conected to a computer, where there is a image processing software installed. One of the cameras is fixed and is recording a scene, for example an empty room. Then a person enters the room, this is what should happen: the image captured from the f ixed camera is sent to the computer, where it is processed and moving objects (the person) are detected - so the image processor generates a X,Y coordinates indicating where is (more or less) the center of that object.
So at this point, the PTZ camera (which has less angle of vision than the fixed one because I have set it to, say, 50% of its maximum zoom) should move to that point the processor indicated. Well, actually it should move to that point having in mind the little difference that exists between the point of view of the two cameras, but that is just a mathematical calculation, not important for this topic.
From the begining I have tried two techniques: first using presets: it would not be a true object tracking, but the PTZ camera would move from position to position trying to follow the object. But the limitation of only 32 presets for a scene of almost 130º horizontal and 80º vertical makes it a no way.
After that I have been "calibrating" the PTZ camera so I calculate my own presets: first measure the time it takes to travel from a fixed point to another, in horizontal movement, and then do the same in vertical. So now when I want the PTZ to go from one point to another, I can "guess" how many time the camera should move in vertical and in horizontal to reach that point. The problem is, maybe you are alerady thinking it, this method is too way inaccurate, because it depends on de TIME (reaction time, the camera seems to be slightly faster when its motors are cold, etc) instead of the distances.
So after trying out ideas 1 and 2, I'm stuck and I need to know if there exists a PTZ control protocol based on distance instead of based on time.