Recognition of a template image from a photo / video from a phone?

Question

Machine learning

Let's say I want to create an application that takes pictures of the desktop FROM THE PHONE, and determines which programs are installed by the icons of the shortcuts. https://prnt.sc/ssrff2

That is, you just need to find the picture in the picture, but not by the screenshot, but by the photo of the monitor. If I made screenshots of a certain size, then I could just use the OpenCV function. cv2.matchTemplate

Let's say there are 50 shortcuts on the desktop, and there are only 500 in the database. white noise may consist.

1. Is it possible to implement such an application offline on an average phone, meizu m6s, for example, through OpenCV, Tensorflow or something else, how many megabytes will it take?

2. If it's fastcan be recognized only in the cloud, how many days will it take to recognize one photo?

Recognition time is no more than 10 seconds.