How to find movement onset after periods of inactivity

I have position vs time data for a subject walking on a treadmill that is moving at constant speed. Most of the time the subject is walking at the same speed as the treadmill, so the position is relatively constant and the speed (in terms of the position) is around zero. Sometimes, though, the subject stops walking and thus moves backward on the treadmill. Before the subject goes off the end of the treadmill, it starts walking again. I would like to extract all the times when this transition from not walking to walking again occurs. Below is a plot of velocity (pixels/second) over time (seconds) with all the peaks I'm talking about manually labeled:

I'm thinking I want to have a sort of moving window looking for periods where the average velocity is below a low threshold for a minimum length of time before then jumping up to an average velocity above a high threshold for a minimum length of time. I would want the times that correspond to the midpoint in the transition from low to high velocity. I'm just not sure how to implement something like that. I'm also working on making the data less noisy both in the position detection (refining the DeepLabCut network) and with filtering, but my main issue is finding these transition times.

edit: attached .m file with signal (100 samples/second)

Answer 1

[Image Analyst]

Is this what you're looking for:

% Demo by Image Analyst
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 22;
s = load('velexamp.mat')
v = s.velexamp;
subplot(2, 1, 1);
plot(v, 'b-');
title('Original Signal', 'FontSize', fontSize);
grid on;
% Smooth the signal with a median filter
vSmooth = medfilt1(v, 151);
subplot(2, 1, 2);
plot(vSmooth, 'b-');
grid on;
title('Median Filtered Signal', 'FontSize', fontSize);
% Set a threshold of 100
threshold = 100;
% Put red horizontal line there.
yline(threshold, 'Color', 'r', 'LineWidth', 2)
highSignal = vSmooth > threshold;
% Find starting points.
startingIndexes = strfind(highSignal', [0, 1])
% Put red vertical lines there.
for k = 1 : length(startingIndexes)
    xline(startingIndexes(k), 'Color', 'r', 'LineWidth', 2);
end

[Steven Lee Ceto]

This does do a pretty good job for a simple approach, but the third peak around 3500 would not meet my criteria because the baseline velocity is too high before the peak. Since the velocity is not corrected for the speed of the treadmill, a velocity near zero means the subject is actually walking. The times that are closer to -200 are when the subject is not moving.

The median filter does do a nice job of accentuating the important peaks, though.

[Image Analyst]

You could of course raise the threshold to not include that one.

You can also change the window size on the median filter to change how "filtered" it looks.

If this code does the trick, then could you click the "Accept this Answer" link? Thanks in advance.

[Steven Lee Ceto]

While the code does appear to mark the desired times in this small snippet, it doesn't really contain the requirement for no movement before the peak that I need. Additionally, the median filter blends some peaks together and results in significant time shifts, like in the two peaks around x = 1000 in the pic below (see attached v_fullrun.m for the velocity over the full trial).

[Image Analyst]

For each starting time, you could examine the filtered time and see if the value is below some "no movement" threshold for the required number of elements.

[Steven Lee Ceto]

Okay, I think with all the tips you've provided I will be able to get something workable. I'll close the question, thanks!

Answer 2

[Image Analyst]

Lots of ways you could approach this.

Maybe try mad() to get the mean or median absolute deviation and then threshold it and call strfind() to look for onsets.

You can use bwareaopen() on the threshold to throw out walking signals less than (shorter) a certain number of indexes.

It would be easier to help you if you would have attached your signal.

[Steven Lee Ceto]

Ah, good point, thanks! I've edited the post to attach the signal.

[Image Analyst]

How many indexes (time span) before a moving/walking period do you consider it to be a still period, and how long does the signal have to be high for the signal to be considered a legitamate motion and not just a noise spike?

And let me know if you have the Image Processing Toolbox because there are some functions in there that will make this a lot easier.

[Steven Lee Ceto]

I would say around 200 samples (2 seconds at 100 samples/second) for the still period and 50 samples for the high signal. Yes, I do have the Image Processing Toolbox.