CODEWITHSUNDEEP
matlabjulia
aberdysh
aberdysh

Reputation: 1664

Estimate normals for point cloud in Julia

Is there an equivalent of MATLAB's pcnormals in Julia?

I also would like to know if there is a way to read and write .ply files in Julia?

Upvotes: 1

Views: 273

Answers (1)

Dan Getz
Dan Getz

Reputation: 18217

From the description of pcnormals here is an attempt to recreate it using Julia with the aid of the Distances and kNN packages:

using Distances
using kNN

function pcnormals(pcloud::Matrix,k::Int = 6)
    n,d = size(pcloud)
    S = pairwise(Euclidean(),pcloud')
    NN = hcat([kNN.k_nearest_neighbors(k,i,S) for i=1:n]...)'
    normals = hcat([normalize(pcloud[[i;NN[i,:]],:]\ones(k+1)) for i=1:n]...)'
    return normals
end

N = 1000
D = 3
X = mapslices(normalize,randn(N,D),2)
normals = pcnormals(X)

println("mean inner product = $(mean(X[1:10,:]*normals[1:10,:]'))")
diag(X[1:10,:]*normals[1:10,:]')

The code after the definition is a mini-test which tries to look at points on the unit sphere and see if the normal of a point is close to itself (using inner-product as a similarity measure). The results looked convincing:

mean inner product = 0.18584539662300542
10-element Array{Float64,1}:
 0.990708
 0.999839
 0.997276
 0.999705
 0.99959 
 0.999883
 0.999052
 0.998935
 0.9951  
 0.999617

So, unless the quick look at pcnormals manual page has mislead me, this may be a starting point for a Julia implementation. Note the use of pairwise in the code which could be slow for big point clouds.

Upvotes: 2

Related Questions