How to detect the change points in a signal?

Question

I have a signal that I obtained by computing the pixel-based-sum of the difference between consecutive image frames in a video. Basically, I want to detect whenever there is a lack of continuity between two scenes. The signal structure seems to change between non continuous frames. However, I am not able to quantify such changes. For instance, in the image below, there is a change at 10450, 11000, 11800, 12,500, 13,500, 14000 etc. How can I detect these points automatically?

Have you tried the changepoint package? https://cran.r-project.org/web/packages/changepoint/index.html — tomaz, Sep 18 '16 at 06:15
Reminds me of the video use case in this article: http://link.springer.com/article/10.1007/s10618-012-0300-z — Has QUIT--Anony-Mousse, Sep 18 '16 at 12:21

score 0 · Answer 1 · answered Sep 18 '16 at 07:55

0

One way would be to calculate the variance and the mean of your samples over a number of data points continuously and to compare the value of this mean and also the variance. By doing that, you will reduce the noise of your data and be able to see clearly a difference. You can afterwards use a Hypothesis test to give a statistical proof of the modification.

answered Sep 18 '16 at 07:55

Crapsy

101

The approach is inefficient for very long signals. – user2808118 Sep 18 '16 at 09:14

How to detect the change points in a signal?

1 Answers1