Hello all,

I have seen various discussion of 'old school' cloud detection using thermopile's vs wider field image capture and star counting.

Back in 2010 I had designed a weather station incorporating a thermopile (TP) so thought I would bring it out of mothballs and integrate it into INDI using indi_watcher_weather on a PI4.
Taking just the PIC controller, TP and adding an ftdi usb serial interface from the following:
Tiny weather station

BTW: indi_watcher_weather is an Awesome easy application capable of being totally data driven! Thank you to the authors!

Now to compare the TP readings (TP is 80deg+ FOV, not a narrow one) with visible clouds I used scripted indi_set/get properties to capture images from a DSI 1 pro outfitted with a 150deg fisheye lens.
An auxiliary program I wrote interfaces to the PIC16f688 controlling the TP to grab it's sensor and case temperature readings

Taking the TP data and stamping each image captured with time and TP/case temp then using ImageMagic converted the FIT's to PNG then again using IM put together a video.

The images were captured roughly every 60 seconds with north to the top and the moon making a long appearance.

In interested, link to video (I do -no- monetization of my channel so any ad's are not for/from or by me)
Sky video with data

To supplement the time stamping there is also a graph of the TP data
Thermopile readout

On the graph there is an indicated time of 4:56:49 in UT, the video is stamped in EST so time would be 11:56 around 14 seconds into the video

Just passing time on these moonlit nights with 30mph winds

Thanks for reading!

Gene

Are you suggesting a large data acquisition campaign to build a ML system that would train on image and use the other sensor as a reference ?
If so this is quite interesting I am going to setup a 224mc based allsky + aag cloudwatcher in the upcoming months.

I was not thinking along those lines but an interesting idea.
I have found cloud detection either via thermopile or semi-narrow field image evaluation to require more than a little 'training'. Seasonal and transparency affects on thermopile and light pollution
vs elevation/azimuth and sky /transparency for image evaluation. The way I use my scope is scripted multiple objects over a section of the sky during an evenings run so basically multiple variables that may/will change for each object.

I think for my use, image evaluation is where I will end up, not star counting but more image average/median background count over the whole image
Or going a narrow field thermopile

Gene