A Novel Unsupervised Method for Temporal Segmentation of Videos

Abstract

In this paper, we aim to address the problem of temporal segmentation of videos. Videos acquired from real world usually contain several continuous actions. Some literatures divide these real-world videos into many video clips with fixed length, since the features obtained from a single frame cannot fully describe human motion in a period. But a fixed-length video clip may contain frames from several adjacent actions, which would significantly affect the performance of action segmentation and recognition. Here we propose a novel unsupervised method based on the directions of velocity to divide an input video into a series of clips with unfixed length. Experiments conducted on the IXMAS dataset verify the effectiveness of our method.