Smoothing with Gaussian kernel

Hello, folks!

I'm trying to create a function which filters raw accelerometer data so that I could use it for Zero crossing. I know that MatLab has built-in functions (gausswin & filter) but I'm willing to create my own because I need it for my Master's thesis.

I found some function named "smooth" here: https://bit.ly/3gsy2cr

I'm trying to modify it the way I mentioned above but my code doesn't as it is supposed to be. I'd appreciate if anybody will take a look at it and give me some tips, thanks in advance!

Attached is the raw data file (.csv)

Here is the code:

clc
clear all

%% Importing raw data
data = readtable('Subject_3_acc_walking_thigh.csv');
%% 
Fs = 50;             % Sampling frequency  
T = 1/Fs;            % Sampling period 
%% Sitting
X = data.attr_x;
Y = data.attr_y;
Z = data.attr_z;
%Calculate time for sitting
L = length(data.id); % Length of signal
t = (0:L-1)*T;       % Time vector
t = t*1000;          % Seconds to Miliseconds
% Sum
y = X+Y+Z; 

yy = smooth(y1,1,25);

plot(t,y,t,yy)
xlim([4500 10000])
xlabel('t (milliseconds)')
ylabel('Acceleration (m/s2)')
legend('Sum','Smooth')


%% Smoothing
function yy = smooth(y,sig,span)
    yy = y;
    L = length(y);

    for i = 1 : L
        
        if i < span 
            d = i;
        else
            d = span;
        end
        
        w = d - 1;
        p2 = floor(w / 2);
        
        
        if i > (L - p2)
           p2 = L - i; 
        end

        p1 = w - p2;
        
        x_val = i - p1 : i + p2;
        kernel = exp(-((x_val-i).^2)/2*sig.^2); 
        kernel = kernel/sum(kernel);

        for j = i - p1 : i + p2
            yy1(i) = sum(y(j)*kernel);
        end

    end
end