The usual way of doing this is to use one of the many existing copy number variant caller. Doing this yourself is both a) complicated to do correctly (the stats for depth per position are complicated by the fact that adjacent position depths are highly correlated since they share reads), and b) already done by someone else.

TLDR: save yourself time and effort by using an existing CNV caller.

I saw you want to get regions that have significantly higher depth than their surroundings; you must get a p-value to show significant as well.

In my opinion, the mostly easy and correct way is using the tool called Genrich, It will calculate a p-value for each peak you got.