Video Boundary Detection

This is a very simple implementation for the algorithm in "Video Segmentation via Temporal Pattern Classification" paper. The kernel being used is the full similarity kernel. The sequentail number of the scene will appear on the upper left corner of the window. If the change in scene is gradual, the symbol '-' indicates that, as in the picture above. You can put any mp4 video inside the directory data for test.

General

• Language: C++.
• Library: OpenCV.
• Build System: CMake.
• Video extensions: MP4.

Build

The project can be built with CMake. Once cloned, cd to the repository (i.e. Scene-Boundary-Detection), and you can follow the below steps:

$ mkdir build
$ cd build
$ cmake ..

Then

$ make

For the following step, the xxx.mp4 is any mp4 file located inside the directory data:

$ ./detectScenes ../data/xxx.mp4

Notes

• This repository is only the test part of the algorithm, the training repository will be uploaded for better training.
• The used OpenCV library has some issues with AVI extension.
• There is some overhead. It is to buffer the computation and show the result immediately, regardless of the video length.
• Performance is an issue right now: no optimization is being utilized.
• Frames are being processed as they are. No change in size, skipping frames or any other kind of downsampling.

Credit

Included videos are from Vimo library. How to download from Vimo: official link.

• Soto Coffee, SotoCoffee.mp4, by Miles Peterson.
• How to Make Japanese Iced Coffee, icedCoffee.mp4, by Counter Culture Coffee.