Evaluating Multiple Kinematic Feactures For Efficient Keyframe Extraction

Abstract

Key frame extraction is a first step in a Video summarization process in order to represent a video with succinct frames after removing redundant frames. Extraction of key frames is a difficult task as key frames changes based on analyst. A simple approach to extract frames is finding of frame difference between the successive frames and then tag as a key frame if the difference is more than threshold. The approach fails when there is a change is actions. In this work, we propose a novel work, which extracts kinematic features that are extracted from the optical flow for human actions in videos. To find out efficient key frame multiple kinematic features are used. And use them as feature descriptor for an image and then applying of fuzzy comprehensive approach with mechanism to find key frames. This helps to find more relevant frames. The work is evaluated on a video dataset and compared the methods based on multiple kinematic features .The experimental results show that our technique gives better results when compared to the competitive methods