What methods to use to recognize the lights on in the opposite house on an android smartphone?

Question

Image processing

The camera for image analysis is only on a smartphone, so I wanted to use the hardware of a smartphone and not transfer the image to a PC The camera lens shows a multi-storey building (partially with tree branches) I wanted to convert the state of an apartment building into a two-dimensional array with two states "lights on" and "off" light"
I consider it as an option in which I manually mark the areas of each window and calculate the average color of each area, as well as a more complicated option, where a long camera adjustment and area matching will not be required.
How computationally difficult is this task for a budget smartphone and does it require the use of a smartphone video accelerator ?
What can you advise to solve this problem?

Answer 1

To solve the problem, I advise you to read the literature on digital image processing. It describes various methods.
The MACHINE LEARNING tag has nothing to do with the solution of the above problem.
Determine where the light is on:
Frame the image to the outline of the house.
Reduce the image resolution to an acceptable level, but not excessively so as not to be mistaken for noise. This reduces the computational work.
Maybe even out the lighting.
Apply Gaussian or median filter.
Convert to b / w image with some threshold - we get white spots of almost rectangular shape on a black background.
Further, object detection - to determine the coordinates of objects.
And everything else is work with an array of coordinates.
The Android app store is full of scanning and imaging apps, including OCR. This means that the computational work is feasible and there is nothing to worry about.