Script for processing video files

It is necessary to develop a script (possibly in Python) that automatically cuts clips from a video based on similarity to clips from another video. The script should take into account possible frame transformations (e.g. crop, scaling, stretching) when comparing.

How it should work: Two video files A and B are input, A - a large file with a long video, B - a short video file consisting of clips cut from file A. The script should find similar video segments in the long video that are in the short one, cut them out, and compile them into a new video C.

In the ideal result, video C should resemble video B but the content is taken from file A.

The quality of video B may differ from the quality of video A, it may contain a logo or the like and be slightly transformed (cropped or stretched), so it is necessary to provide the ability to adjust the similarity threshold to determine similar scenes.

How I envision the process (you can suggest your own):

1. Extracting frames from both videos

Input data: two video files, Output data: Saved frames in image formats (JPEG/PNG) in separate directories.

2. Comparing frames.

(Input data: Directories with extracted frames from both videos.

• Process: Comparing frames using alignment and keypoint algorithms (e.g. ORB) or your own variation.
• Considering transformations: The script should take into account possible crops, scaling, and stretching of frames or your own variation.
• Similarity threshold: Ability to adjust the similarity threshold to determine similar scenes.
• Output data: List of pairs of similar frames with their timestamps.)

3. Video cutting

• Input data: Original video and list of pairs of similar frames with timestamps.
• Process: Using FFmpeg (or equivalent) to cut the output video based on the timestamps of similar frames.
• Output data: Saved video clips.

Requirements:

The saved video clip may differ from video B but not by more than 80%, for testing I will provide 3-4 examples.

Technology requirements: whatever is convenient for you, but it seems that you can use:

• Python
• Libraries: OpenCV, NumPy, FFmpeg
• Preferably make the script cross-platform (Windows, MacOS)