How can I integrate vision based logic such that I can tap/swipe based on visual match instead of UI element ?

[miclejackson12]

I m traking about how similar to ui.vision is accessing moble via anydesk in this video

https://youtu.be/wH0MLIUdD-Q detailed tutorials here

I was wondering what would it take to do the same using uiautomator such that I dont need the anydesk in between.