Change of variables to flatten the boundary

Question

It is known that one can perform a change of variables to flatten a $C^2$ domain $\Omega$, that is, for any point $x \in \partial \Omega $, there is a $C^2$ diffeomorphism $\psi$ which maps a neighbourhood of $x$ in $\Omega$ to the upper half ball $\{x \in B_{1}(0): x_n>0 \}$. I can't see this, a reference will be great too. Thank you.