Human detecting Sensor

[Ŕähûł Śïñģh]

trying to make human detecting sensor which can detect the presence of human automatically.

[marvin78]

How do you plan to do that? Compnonents?

[wallyllama]

@ŕähûł-śïñģh if you are successful you'll be rich. There are simple detectors like pir, and complex like cameras with computer vision software. Im working with amg8832 chips with 8x8 grid of temp sensors, essentially a low res camera. I looked at radar "pir" modules and thought modulating the voltage would give data that can be processed, again complex, combining pir and ir door sensors has been suggested. None of this is my original idea, im just trying to piece something together.

[Newzwaver]

Any ideas on how to start? Maybe we can all help.

[dbemowsk]

@newzwaver have a look at this thread that I started a while back:
https://forum.mysensors.org/topic/7814/a-smart-home-vs-an-automated-home
there is a lot of discussion on this topic.

[Constantine Poltyrev]

@wallyllama Just got my amg8832 evaluation board. Do you have any presence detection code to share?

[wallyllama]

@constantine-poltyrev i dont have code yet. I havent really started on software yet. My thinking is to start by using the adafruit code as a base, set up the sensor in a likely spot and modify the code to stream data to a pc or r-pi, maybe with a web cam next to it for comparison. Id look to see if there are patterns visually that are helpful, but my real hope is that since the pixels are temperatures, the ones with people are at people temperature, so around 37°c.

Maybe the first filter would call out pixels in the 35-40 degree range, and send 0 for the rest. If that gives a spot that has few false readings it may be good enough for "there is at least one human present" or not readings. A glorified pir.

Likely though a person will be more than 1 pixel, or worse part of more than one pixel, how to know if two cooler pixels represent 1/2 a person each, then there is the problem of the sun heating something to human temperature. My gut feeling is these are simple enough that existing solutions can be adapted. I dont know what those are yet, maybe bayesian filters, maybe some sort of computer vision.

A hardware product called openmv may provide some inspiration. It uses something called micropython and opencv. With 64 pixels we have a simplifies case, but I think an arduino is not going to be enough, a raspberry pi or maybe a nanopi neo should be sufficient.

I think that is the current contents of my brain on this, experiments are the next step. Starting with streaming some data and looking at it.

[Constantine Poltyrev]

@wallyllama I experimented a bit with the software that goes with the evaluation board. It looks like the temperature filter should be dynamic as the measured temperature for a single pixel depends on object distance since each pixel has a view angle and the resulting temperature is an average of the covered area which obviously grows with the distance. So in real life you get 24 to 31 Celsius for people in the room if you have the sensor mounted in the middle of the ceiling.
Panasonic has some people tracking demo app and they even provide it's source if you sign their SLA, so it also could be a good starting point. I haven't obtained the source yet though.