Convhull function returns indices that correspond to NaN values

I'm trying to calculate which points contribute to the convex hull of a movement trajectory. Coordinates of the movement trajectory are organised in a matrix (10000x360). Each colom is a different trajectory, each row is a data point. The trajectories all have a different length, so when there is no data anymore for that colom, the rest is notated as NaN's.

For some trajectories this works really well, but also for some I get an indices which corresponds to a NaN value. So the convex hull uses a point that does not exist. How can this happen and how to prevent?

for i = 1:360 % 360 trajectories
    m = Indices(i); % Indices gives the last value which is not a NaN of each trajectory (i). 
    P = [x(1:m,i),y(1:m,i)]; %combining x and y, of not NaN values.
    indicesCHull =convhull(P);
    XCHull =x(indicesCHull);
    YCHull =y(indicesCHull);
    % more following here, but not relevant I guess
end

[Image Analyst]

Attach the (10000x360) matrix in a .mat file.

If you have any more questions, then attach your data and code to read it in with the paperclip icon after you read this:

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

Answer 1

[Moksh]

Hi Sanne,

I understand that you would like to create a convex hull for each trajectory represented by columns in the “x” and “y” matrices. Since matrices in MATLAB require fixed column sizes, you have filled the empty entries with NaN values.

Instead of maintaining a separate index array to track non-NaN values, you can utilize a “cell array” in MATLAB. Cell arrays allow for storing variable-sized vectors. By using a cell array, you can apply the "convhull" function to each trajectory without generating any NaN values.

Here is an example code for how you can convert this 2D matrix to a cell array:

% Initialize a cell array for maintaining clean x and y trajectory values
XcellArray = cell(1, size(x, 2));
YcellArray = cell(1, size(y, 2));

% Iterate over each column
for col = 1:size(x, 2)
    % Extract the x, y column data
    XcolumnData = x(:, col);
    YcolumnData = y(:, col);
    
    % Remove NaN values from x and y columns
    XcolumnData = XcolumnData(~isnan(XcolumnData));
    YcolumnData = YcolumnData(~isnan(YcolumnData));
    
    % Store the column data as a vector in the cell array
    XcellArray{col} = XcolumnData;
    YcellArray{col} = YcolumnData;
end

% Now proceed with the convhull function on clean trajectory values
for col = 1:size(x, 2)         
    x_traj = XcellArray{col};     % Extract clean x and y trajectory values     
    y_traj = YcellArray{col};
    P = [x_traj, y_traj];         % Combining x and y, of not NaN values.
    indicesCHull =convhull(P);
    XCHull =x_traj(indicesCHull); % Populate the convhull coordinates on the clean trajectory values
    YCHull =y_traj(indicesCHull);
    % Following logic
end

In the above code, I have assumed that whenever there is a NaN value in the x column, there is also a corresponding NaN value in the y column, and vice versa. If this assumption does not hold, and there are instances where NaN values exist in one column but not the other, you will need to take the intersection of the indexes in both columns that point to the non-NaN values.

Please refer to the below documentation for more information about “cell-arrays” and “isnan” function:

• https://in.mathworks.com/help/matlab/ref/cell.html
• https://in.mathworks.com/help/matlab/ref/isnan.html

Hope this helps!

Best Regards,

Moksh Aggarwal

Answer 2

[Bruno Luong]

You must index the column where the convhull is inquired

XCHull =x(indicesCHull,i);
YCHull =y(indicesCHull,i);