Originally posted by pingflood You pick a focus point, usually the center point, as the starting point. You place your subject (e.g. a bird or a soccer player or something) there and tell the camera to start tracking. Now when the subject moves across the viewfinder the camera will automatically switch which AF point it uses to keep focus on the subject. Soccer player runs to the right, the camera will start switching to the AF points to the right of center, etc. Then there are algorithms and settings that determine how long it will follow the predicted path of the subject before giving up, so if your bird flies behind a tree or a telephone pole, or your soccer player runs behind someone on the field, the camera will keep trying to predict where he/she was headed so that when they reappear it can reacquire focus.
Unfortunately, this could only work for an isolated subject on a simple background. But for this to work reliably with background and foreground distractions there should be much more information available to that algorithm then AF system can provide.
I feel that I can safely make this claim because several years ago I was experimenting with data available in the AF system and provided by the lens in order to devise some sort of software-based focus limiter (especially for macro and telephoto lenses). I gave up eventually and one of the main reasons was because available information was not consistent between the lenses (plus several other problems). But this is what I have learned:
Each individual AF point is actually not a point, it's an AREA and there is no way for a relatively large area to reliably track any small object. Next, distance information is NOT associated with individual AF areas, there is just in/out of focus status. Overall distance scale is transmitted by lenses but it is very rough (like "closeup", "near", "medium", "far" and "infinity" if I remember well). In addition, one additional piece of information could be determined in AF-C mode on how correction is acquired, say, if reacquiring AF with new status "far" required forward correction "medium-to-far" or backward "infinity-to-far". And that's pretty much all.
Although that is enough information to track a subject moving towards or away from the camera or across the frame with plain background, it is absolutely NOT enough to track subjects moving across the frame with background or foreground distractions. The functionality as described in your post would require either pattern analysis or more accurate AF distance associated with each AF point, both not available in any AF system in consumer cameras today.