3D plotted shape is not following curvature of data points

I am writing a code that creates a 3D shape from a set of data points and am running into this problem where the shape does not properly follow the curvature of the data points and rather draws a straight line. This can be seen on the sides of the figure below where the shape isn't properly following the data points into the curves.

My code:

load('seg_info.mat');
x = seg_info(:, 1);
y = seg_info(:, 2);
z = seg_info(:, 3);
colors = seg_info(:, 4);

scatter3(x,y,z);
hold on;
DT = delaunayTriangulation(x, y, z);
[K, v] = convexHull(DT);
trisurf(K, DT.Points(:,1), DT.Points(:,2), DT.Points(:,3), ...
	'FaceAlpha', 0.5, 'EdgeColor', 'none', ...
	'CData', colors, 'FaceColor', 'interp');

Does anyone know how I can improve my code to better fit these sides? I have tried using alpha shapes and boundary in an attempt to derive the concave hull but neither worked.

[jonas]

Could you upload the data?

[Sumeet Chaudhari]

I have attached the data

[Star Strider]

I suggested the boundary function.

Sumeet Chaudnari replied:

‘I tried using boundary with a shrink factor of 1 (which should provide the tightest bounds for the shape) but still end up with the same incorrect shape.’

So I deleted my Answer.

[John D'Errico]

You think the problem is in the triangulation? A delaunay triangulation trinagulates the entire region within the convex hull of the data. In fact, you even used a convex hull there. So surely you would expect that the result will be a convex region. If you don't want it to be convex, then you can't use a convex hull. Or for that matter, a Delaunay triangulation.

[Sumeet Chaudhari]

I have been trying to use mesh and even searching File Exchange for a program that calculates a concave hull. Looking at a scatter3 plot, I wish there was a way to just connect all the dots since the input data already represents the outer surface of the shape.

Answer 1

[jonas]

Try this FEX function

https://se.mathworks.com/matlabcentral/fileexchange/63730-surface-reconstruction-from-scattered-points-cloud

[Sumeet Chaudhari]

thank you so much for the help! really appreciate it

[John D'Errico]

Assuming that is a CRUST implementation, it is probably the best choice. And a quick glance at the code suggests it is indeed CRUST.

An alternative code can be found here:

https://github.com/Daniel-Sanche/Power-Crust-MATLAB