Darshankumar D.Billur¹,Dr.Manu T.M²,
¹KLE College of Engineering & Technology, Chikodi,India, ², KLE Institute of Technology, Hubli, India

Extracting relevant points of action from video sequences has found its applications in both commercial and non-commercial scenarios. Applications like CCTV monitoring, video summarization, video codec optimization, etc. make use of key frame extraction (KFE) to function effectively. KFE has also found its applicability in military applications where military training footages are given for KFE and the output is used for fast track training of the officers. In this paper, we propose a novel hybrid motion vector based KFE algorithm, that utilizes motion vector information and combines it with a multi-color space visual attention model to extract key frames. The proposed algorithm can improve the precision, recall and f-measure values when compared with state-of-the-art algorithms like Delaunay clustering, VSUMM, DT, OV. It is found that the proposed algorithm improves the efficiency of KFE by more than 20% across different video datasets.

Key frame extraction, summarization, multi-color space, motion vector