real world object distance

hello,

I am developing an application that tracks an object via camera (real object, using AR), and detects the distance it across.

for each frame, I calculate the distance between the current position and the first position.

I got wrong position details, and camera movements affect the result.

any solution?

for more clearing I want to implement this experience: (video):

https://youtu.be/-c5GiXuATh4