A Scene Boundary Detection Approach Using Audio Features

Scene boundary detection is essential in video intelligence, video analytics, and video summarization. Cutting long videos into short meaningful and semantic scenes helps in applying video classification tasks and extracting useful information from the video. It’s a challenging task to extract meaningful scenes from the video. Previous work involved scene detection using camera angle change detection and subtitle text timestamps. This approach extracts tiny scenes that are irrelevant in many video intelligence applications that involve extracting long scenes such as video summarization video analytics and video content moderation. Here, we proposed an efficient scene-extracting mechanism using audio features that extract long and relevant scenes by identifying changes in audio features like speech, music, background audio, human voice, animal voice, etc. The proposed approach is efficient in terms of speed and extracting well-formed scenes.