Discrete first and second fundamental forms

Question

I want to calculate first and second fundamental coefficients for some points in a point cloud (sampled surface), I used this method (https://arxiv.org/abs/1601.07272) but in this, only three closest points are used for this reason, so the error is high and it's very sensitive to noise and occlusion. Also I used surface fitting (polynomial with degree 2,3 ,...) then I calculated these coefficients, but the error is high. I searched a lot for finding discrete methods but I couldn't find any good method. Actually I am searching for a discrete method that uses a set of neighbor points for achieving these coefficients.
any help? Thanks in advance