For two video, the content source may be the same, but the start time of the collection is different, or video format difference, or acquisition of signal transmission damage rate is different.
In this case, how to align them both based on video frame and find a common datum point is a problem of the study of a value.
This paper provides a technical solution to this problem, and the author has verified the feasibility:
1, the video is marked as a, B, respectively
2, from the middle of B video extract a paragraph, all the solution frame image Group X
3, all the frames of a will be extracted image group Y
4, the x, y each frame quantified into a vector (here I use the 48-D chroma) for image similarity judgment. The concrete realization Way please consult the graph science related paper.
5, x formation of the vector group to take n points of mutation (adjacent two frames the largest gap, the frame of the gap using a vector model to measure)
6, on the Y form based on the N mutation point of the sliding window, calculate the n point of the Gap, take the minimum sliding position
7, the position is aligned point
A bit abstract, the relevant code for confidentiality considerations, not shown here.
This article does not discuss the issues addressed:
1 A, B frame rate is different (but the basis of this article is very good to modify, as long as the vector takes into account a comprehensive frame distance similar to take the form of "integral")
2 algorithm optimization (on a can not take all the frames, the general business can be estimated alignment range)