Any approach based on thermal/IR sensing would be subject to the sort of problems you mention. However, as one alternative, I think there might be more mileage to be had in RF (probably 'microwave') approaches, which could probably be persuaded to detect when the fields within the room were disturbed by the presence of the mass of a human body ... but that's just an initial thought 'off the top of my head'.
Yes, and I happen to have one to play about with in my workshop. A different principle to thermal sensing, but the net effect is exactly the same - they depend upon a change to detect a reason to switch. The big difference is it needn't be a warm body, just a body, or an item within it's detection range.