The IR tracking that zoomkat suggested sounds like the best option, the people that will be tracked are going to be going in a predictable fashion so losing the signal wont be much of an issue, the jerky movement can likely be smoothed out.
It might work; but let me give you a scenario. Let's say for the sake of argument, the IR beacon the user is wearing is mounted on a tie clip. They stand there, the camera is on them, then they turn to walk left...
Which direction should the camera move? How does it know they turned and are walking left? Do you put another beacon on each arm? Do you put an omni-directional beacon on their head (or make them wear a hat)?
Look at it this way, too: The person is facing the camera, the beacon is too. They side-step to the right 10 feet. How does the camera know to track them in that direction? Sure, the signal from the IR beacon gets less, but it would do the same whether they moved left or right. Do you just take a guess? If you always know what direction they will move before they do, then you could use this knowledge to move the camera properly, but humans being human means that your well-laid plans will likely fail if using this model...
You could put a compass on the beacon, and based on the direction the person was facing, change the signal the beacon put out; however, you would still face the possible occlusion issue should they block the beacon with their body as the move/turn/etc.
These are issues faced every day by motion capture artists and professionals; if it was just as simple as sticking a simple beacon and easily following it, these systems would cost a heck of a lot less. This is why with inbound camera-based systems, there are always multiple cameras on the actor-in-motion, in order to catch occlusions (even so, there is a lot of data cleanup to do afterward).
If you can think of a way (that is acceptable to everyone involved) to put beacons everywhere on the person being tracked (to prevent occlusion issues), and also somehow be able to track direction of motion from these beacons using a simple system on the Arduino controlling the camera, you may have a lot of people beating down your door. I don't think it is going to be as easy, though, without using some kind of camera system (or at a minimum, a single-line CCD imager array, or maybe a linear array of IR detectors) and a lot of beacons "stuck" everywhere on the person's body.
While what I suggested before is a lot more complex than what you were probably wanting, it also probably wouldn't suffer from any interference effects, it would be small and easily concealable (as concealable as a wireless microphone), it could be used outdoors, and it would likely be fairly accurate.