Post by impactoz on Sept 6, 2019 15:48:05 GMT -8
Before I embark on my next enhancement - seeking feedback of thoughts...
I have been googling for awhile in how to determine cloud coverage... Originally started thinking about identifying clouds and maybe using deep learning to identify the types of clouds - cumulus, cirrius etc. But certainly moved away from that idea - as all I can find via google is 1 or 2 sites that gives the indication that not even supercomputes can detect the types of clouds properly. (it is so hard to find info as it is always imbedded in cloud computing - arghh why could they not name cloud computing something else - I don't want to remotely save files somewhere!)
I appear to have 2 options - use a thermal IR thermometer to detect sky temperature and compare with the ambient temperature. The difference indicates how clear the sky is apparently... So I have bought a detector for that - but thinking I may not go that way - concerned that the wind vane / anemometer might get in the way of a good reading...
Or I use analysis on the pictures taken... Thinking I will go that path and count the number of blue pixels, or range of blue pixels in the SkyWeather.jpg file....
Found that there is a standard way of determining overcast / clear days.... Its called Okta -> en.wikipedia.org/wiki/Okta and the values are clear (0); Few (1 to 2); Scattered (3 to 4); Broken (5 to 7); Overcast (8).
So now thinking of using a scale to determine the number of blueish pixels in the SkyWeather.jog to determine how bright it is, subtracting that from the total pixels of the picture (less any masking, if you had buildings / trees in view) - that would then give you a variable which you can use to say whether it was currently clear or overcast.
Or I may just count white / grey pixels...
I would probably save the values in the db, so that you could get an average for the day to say the day was mostly overcast or something...
Whats your thoughts?
My first stumbling block is that SkyWeather is currently running on Python 2, and OpenCV requires Python 3, so I am concerned about compatibility of doing that python upgrade. Yes I can run both, but I was going to embed this pixel count inside the TakeSkyPicture routine, so that's a Python 2....
SDL - Any experiencing in upgrading the software to Python 3 Or are there reasons its currently on Python 2 ?
I have been googling for awhile in how to determine cloud coverage... Originally started thinking about identifying clouds and maybe using deep learning to identify the types of clouds - cumulus, cirrius etc. But certainly moved away from that idea - as all I can find via google is 1 or 2 sites that gives the indication that not even supercomputes can detect the types of clouds properly. (it is so hard to find info as it is always imbedded in cloud computing - arghh why could they not name cloud computing something else - I don't want to remotely save files somewhere!)
I appear to have 2 options - use a thermal IR thermometer to detect sky temperature and compare with the ambient temperature. The difference indicates how clear the sky is apparently... So I have bought a detector for that - but thinking I may not go that way - concerned that the wind vane / anemometer might get in the way of a good reading...
Or I use analysis on the pictures taken... Thinking I will go that path and count the number of blue pixels, or range of blue pixels in the SkyWeather.jpg file....
Found that there is a standard way of determining overcast / clear days.... Its called Okta -> en.wikipedia.org/wiki/Okta and the values are clear (0); Few (1 to 2); Scattered (3 to 4); Broken (5 to 7); Overcast (8).
So now thinking of using a scale to determine the number of blueish pixels in the SkyWeather.jog to determine how bright it is, subtracting that from the total pixels of the picture (less any masking, if you had buildings / trees in view) - that would then give you a variable which you can use to say whether it was currently clear or overcast.
Or I may just count white / grey pixels...
I would probably save the values in the db, so that you could get an average for the day to say the day was mostly overcast or something...
Whats your thoughts?
My first stumbling block is that SkyWeather is currently running on Python 2, and OpenCV requires Python 3, so I am concerned about compatibility of doing that python upgrade. Yes I can run both, but I was going to embed this pixel count inside the TakeSkyPicture routine, so that's a Python 2....
SDL - Any experiencing in upgrading the software to Python 3 Or are there reasons its currently on Python 2 ?