Yes, I guess I need something that sees the heat.
After, in the case of the TV room, I will plan to also have something like, if TV ON, dimm down the light etc. But not all my rooms are tv room, second, I can go to the TV room without watching TV.
I will try the camera and motion detection.
If you always carry your phone, or something else “detectable”, you could also weave this into the presence detector.
Depending on the kind of sofa, maybe some body sensors/switches as well, similar to what they have in elderly home. But I don’t know if there are any low cost ones. Maybe it is possible to make one yourself?
Also the IR transmitters/receiver pairs used for garage doors might do. (the one where you break the IR light) But it may not look to nice…
The problem with PIR sensors is that they have problem picking up small movements. According to this blog (in Dutch) the B+B HYG RAD-MOD radarmodule picks up small movements much better than a PIR. The module is available from amongst others Reichelt and Conrad but I’m sure they can be bought at other places as well.
I have no idea how well they work. I’m more a software guy than an electronics guy. I guess you should be able to somehow hook them up with an Ardiuno, ESP8266, RPi, etc. and then trigger an event in openHAB using the MQTT protocol.
I would really suggest you reconsider what @rlkoshak said about diminishing returns.
Let say you have 14W CFL lightbulb (leds are even better), running 24h/7 days a week, with 0.1$ price per kWh (adjust as needed).
That totals to 12.27$ per year. Your IR sensor costs around 50$ (radar module even more), which means you are not going green at all (consider manufacturing costs, running costs etc.). So having a perfect solution is much less Eco friendly as having a good enough solution. That is also an important engineering lesson.
EDIT: cheap microwave sensors mentioned by @marcel_erkel could be game changer, but I have not tested them yet, so I hold my opinion for now.
However if we say you are not aiming to go green, but want 100% solution no matter the costs, I suggest you go with combination of several sensors coupled with rules evaluated by learning neural network, which will get better over time.