Find time of slope change

I want to find the time where the slope of the graph starts to decrease. In this case, the graph is a measurement of the foot angle of someone walking. I would like to use MATLAB to detect when the person starts walking, which in this case seems to be at around 2 seconds. The problem is that there is a dip in the graph right at the beginning. Is there a way to detect an area of a graph with a slope close to 0, and then find where the first change that happens after that? I would like to also be able to use this code for other sets of data that do not look exactly the same.

Answer 1

[jonas]

If you have the signal processing toolbox, you can try the function findchangepts. I believe the very similar ischange would be equally useful.

Here is an example. See attachment for results.

pts=findchangepts(y,'Statistic','linear','MinThreshold',20);
      plot(x,y,'-b',...
          x(pts),y(pts),'rx')

[Sarah Levovitz]

Hey Jonas, thank you very much for your fast reply. Okay so yes, I understand that I could find any changes in our data. However, the change I would be looking for is the one that is after the flat line. Of course I could just take the second change. However, this code is made to analyse different data every time. The main shape of the graph will look the same, although maybe next time I take measurements there wouldn't be a peak initially. The ideal thing would be that I wouldn't have to look at the graph each time to check if I should take the first, second or third change. The way to do it is check when there's a flat line continuously, and then there's a negative gradient. I just don't know how to do this...

[Image Analyst]

Can you attach a picture of where you want the points to be found? Also, attach it as a picture we can see in the browser without having to save a .fig file, switch to MATLAB, then open the figure. A screenshot like Jonas did is more convenient for us to see it.

[jonas]

The first thing you should ask yourself is: how would you, qualitatively, detect the correct peak? In other words, what characterizes the point of interest ? From what I can understand, it the end of a long plateau.

If so, then it is quite easy to find it. For example, we can note that the segment that comes before the point of interest is, by far, the longest segment. We can find it as follows:

[~,ind]=max(diff(x(pts)))
start=x(pts(ind+1))

The first line outputs ind=1, which means that the point you are looking for is the point at the end of the first segment, i.e. x(pts(ind+1)).

Now, is it possible that there are other special cases when you will an even longer plateau somewhere else, e.g. at the end of the time-series? If so, then perhaps we need to find the earliest segment which is longer than a certain threshold.

Anyway, only you know what works for your data!

[Sarah Levovitz]

thank you very very much, this worked perfectly!

[Sarah Levovitz]

@Image Analyst, thank you for the comment, I will note this for future.