Best way to filter a Matrix

I have the postion, velocity and acceleration data from a subject walking. Each variable is a matrix with each column representing a step and each row representing a measurement. However, there are some bad values where VZ didn't record a position (ie 1; 2; 3; 0; 5) or the vaue is way off(ie 1; 2; 3; 1256; 5). I would like to create vectors for average step values for position, velocity and acceleration.

Whats the best way to create the vectors without the bad values?

I am having trouble with the finding the bad values and making the dimensions of the columns agree. Any help is greatly appreciated... Thanks.

Answer 1

[Paulo Silva]

Maybe this can help you start with a smooth algorithm

po=[1 2 3 10 4 5 100 6 7 8 9 0 10 11 12 0 13 14 15 16]
p=po;
v=diff(p);
vp=find(abs(v)>2);
vp=vp(1:2:end);
p(vp+1)=(p(vp+2)+p(vp))/2;
t=1:numel(p);
plot(t,po,t,p)
legend('Measurements','Filtered measurements')
p

[Daniel]

I'm working with your code to see if I can make it work.

http://i3.photobucket.com/albums/y80/griffdrc/Graph.jpg

I need to create graphs like Figure C and D. Thank you for your help!

[Daniel]

Thank you for yor help... I was unabe to replace bad values with averages. I placed limits on the variables and it worked okay.

[Paulo Silva]

I just remember something that might be useful for the same purpose, it's the rate limiter from simulink, http://www.mathworks.com/help/toolbox/simulink/slref/ratelimiter.html , you can use the documentation formulas on your problem.

Answer 2

[Sean de Wolski]

Do you have the curve fitting toolbox? If so, look at:

doc smooth

specifically the 'rlowess' option.

Are all three signals recorded or do you just have acceleration or just position? Assuming you only have one of the signals, do the smoothing before any differentiation or integration.

[Daniel]

I do not have the curve fitting toolbox...

I have position, velocity and acceleration data. I was not present at the recording. When I try to filter the data before other calculations (sparating the data by step and task) it takes longer and I run into the same problems.

Answer 3

[Andrei Bobrov]

my variant

pf = [po(1)-1 po po(end)+1];
p = median(pf(bsxfun(@plus,1:3,(0:length(pf)-3)')),2);

[Daniel]

I am using your code and the code posted above. It is tricky because there can be multiple consecutive bad values.

http://i3.photobucket.com/albums/y80/griffdrc/Graph.jpg

I need to creat graphs like the ones in Figure C and D.