I'm working on the idea of recognizing a small video sequence (a sequence of 18 seconds) within a long one (a sequence of 15 minutes), the first idea that come to mind stats that I have to characterize this small video with some kind of identification. so I will use the identification to look this small sequence inside the long sequence, I'm asking this question to the community, to share with me an approach that can help me in characterize this small video, then recognize it within the long sequence.


  • $\begingroup$ Welcome here! You've sadly spent all your question's body to repeat the sentence in the title (with minor modifications), but we're probably more curious about what you know about the videos, what "small" and "long" are (roughly), the resolution of the video, and what kind of other restrictions you have on your problem! In general: Try to give as much background as feasible, so that anyone willing to answer knows what you need instead of only what you directly ask for. $\endgroup$ – Marcus Müller May 22 '18 at 17:47

The most intuitive approach here would simply be taking every pixel's color values, and finding peaks in the individual pixels' reference video/short video cross correlation function. That's basically a W×H-dimensional crosscorrelation function; you could speed that up by doing fast convolution.

You could algorithmically reduce the effort by first reducing the dimensionality of your problem, for example by reducing each video frame to a global brightness value and a mean motion vector of e.g. the frame center region – then you'd only have to calculate a threedimensional crosscorellation function.

| improve this answer | |
  • $\begingroup$ thank you, Sir, I will implement this approach, however what about the complexity of your idea? is it too high ? $\endgroup$ – azdoud May 23 '18 at 17:09
  • $\begingroup$ I don't know what "too high" means, because you don't give any constraints on that. I already recommended a reduced-effort approach, so I'm not even sure why you ask this. $\endgroup$ – Marcus Müller May 23 '18 at 17:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.