Removing negative values of 3d data

Hello, I have a diagram where I want to get rid of the lower portion (the points which have negative z value) so I can use scatteredInterpolant on it. How would I do this?

The points are stored in a 3D array with index (1,1,1) being an x-value, (1,1,2) being the corresponding y value and (1,1,3) being the corresponding z value.

Thanks!

[Torsten]

And what are the first two dimensions of the 3d-array for ?

[Image Analyst]

Do you truly have a 3-D, volumetric array, or do you actually have an N-by-3 two dimensional matrix with each row being the x, y, and z values? I would think it's the latter. What does the size() function report for your array?

[Nathan Mu]

The array is 357 x 177 x 3

The first two dimensions are to do with different points ?

Actually the graph i have here is the original surface rotated onto its side (I wanted to change the input variables). The original surface had the circular shape on the left on the xy plane.

[Torsten]

After removing the data with negative z-values, you will have to work with a 2d-matrix, I guess.

A = reshape(A,[357*177,3]);
idx = A(:,3) >=0 ;
A = A(idx,:)

[Image Analyst]

So you have a true 3-D array (though I'm still very doubtful). How are you plotting the surface then? Are the "invisible" voxels NaNs then? Can you attach the array in a .mat file, after you read this:

TUTORIAL: How to ask a question (on Answers) and get a fast answer

@Torsten, no, if it's a true 3-D matrix, then moving any plane/slice that has a negative value will still leave you with a 3-D array. Just think about it. If he had a volumetric image and one big spike went below the x-y plane, and you somehow set that value to another value or nan, you'd still have the same 3-D array, just (perhaps though not guaranteed) with fewer slices.

I'm still thinking he really has a 2-D array numPoints-by-3 matrix with the first column being the x values, the second column being the y values and the third column being the z values of the surface of that shape he's showing. Just because you have 3 separate columns for x, y, and z does not make it a 3-D array. What does

arraySize = size(your3DArray)

show?

[Torsten]

@Image Analyst

If he had a volumetric image and one big spike went below the x-y plane, and you somehow set that value to another value or nan, you'd still have the same 3-D array, just (perhaps though not guaranteed) with fewer slices.

Yes, but also 3d-data can be saved in a 2d-matrix. It's only uncomfortable to access the data of interest in general.

The question in my opinion (and also in yours, I guess) is: what do the first two dimensions of the matrix represent ?

[Image Analyst]

Exactly. That's why I'm thinking he really has a numPoints-by-3 two-D matrix. Of course if it's a 2-D matrix you can just record the presence or absence of a surface, not a value for the voxel like you'd get with a true 3-D image, like from CT or MRI.

[Nathan Mu]

I've attached the array here. size(GGV) reports 357 117 3.

GGV(:,:,1) is the table with the longitudinal acceleration (ax) values, GGV(:,:,2) is the table with the lateral acceleration (ay) values, and GGV(:,:,3) is the table with the speed (v) values.

The original (non rotated) surface looks like this:

which was plotted as

surf(GGV(:,:,2), GGV(:,:,1), GGV(:,:,3))

I am plotting the rotated surface with

surf(GGV(:,:,2), GGV(:,:,3), GGV(:,:,1))

which has lateral acceleration as the x-axis, speed as the y-axis and longitudinal acceleration as the z-axis.

[Nathan Mu]

Actually, I think i may have some idea of what to do. I have to somehow restructure the GGV array into a 2d array with 3 columns (x,y,z) like you were saying, and then it should be pretty easy to remove the unwanted points, right?

And reshaping it makes the plot look like this somehow:

[Torsten]

This answer should be helpful in your case:

https://de.mathworks.com/matlabcentral/answers/384291-3d-surface-plot-out-of-points

Or you could just replace the data points with negative z by NaN - they will not appear in the plot afterwards.

[Nathan Mu]

Okay thanks!

Answer 1

[Chunru]

x = load("ggv.mat");
GGV = x.GGV;
figure
surf(GGV(:,:,2), GGV(:,:,3), GGV(:,:,1), 'EdgeColor', 'none');
xlabel('x'); ylabel('y')

figure
% to demonstrate limit works for X data (Z in the plot)
surf(GGV(:,:,2), GGV(:,:,3), max(GGV(:,:,1), -20), 'EdgeColor', 'none');  % -20 and above
xlabel('x'); ylabel('y')

% The number of points for z<0
sum(GGV(:, :, 3) < 0, 'all')

ans = 0

min(GGV(:, :, 3), [], 'all')

ans = 0

figure
% limit z data here will not make no change since no z is less than 0 
surf(GGV(:,:,2), max(GGV(:,:,3), 0), GGV(:,:,1), 'EdgeColor', 'none'); 
xlabel('x'); ylabel('y')

figure
% limit GGV(:,:,1) 
surf(GGV(:,:,2), GGV(:,:,3), max(GGV(:,:,1), 0), 'EdgeColor', 'none'); 
xlabel('x'); ylabel('y')
zlim([-80 20])

[Nathan Mu]

Sorry, I forgot to mention that I am trying to remove the points where GGV(:, :, 1) < 0.

[Chunru]

See above update.

[Tim Albertsson]

Hi! How did you get the GGV points from the start? I'm trying to make my own script for this and I've gotten the max lat and lon accelerations for the different speeds but not the full (GG "elipse").

Answer 2

[Chunru]

% Your data
data = randn(357, 177, 3);

% limit to 0
data(:, :, 3) = max(data(:, :, 3), 0);   % to limit the surface up to 0

% You may also want to try nan (depending on how you plot the data)
idx = data(:, :, 3) < 0;
data(idx, 3) = nan;

[Nathan Mu]

Yep I have tried something similar to this. But I want to be able to use scatteredInterpolant on this so unfortunately it doesn't work.

[Chunru]

Then why does it not work?

[Nathan Mu]

I wanted to get rid of the points which have z<0 so I could use it as a function. So I would input lateral acceleration and speed and it would give out a longitudinal acceleration. However with the GGV array as is, there are two different values of long acc for any given lat acc and speed, so it isn't a function which can be put into scatteredInterpolant.

The issue may actually be due to how the array is structured. So I'll try to restructure it as a 2D matrix with three columns (x,y,z) and see how that pans out.

[Chunru]

The number of points for z<0 is 0. So you cannot see any difference after limiting

sum(GGV(:, 3) < 0, 'all')