Fusion of foreground object, spatial and frequency domain motion information for video summarization

Abstract

Surveillance video camera captures a large amount of continuous video stream every day. To analyze or investigate any significant events from the huge video data, it is laborious and boring job to identify these events. To solve this problem, a video summarization technique combining foreground objects as well as motion information in spatial and frequency domain is proposed in this paper. We extract foreground object using background modeling and motion information in spatial domain and frequency domain. Frame transition is applied for obtaining motion information in spatial domain. For acquiring motion information in frequency domain, phase correlation (PC) technique is applied. Later, foreground objects and motions in spatial and frequency domain are fused and key frames are extracted. Experimental results reveal that the proposed method performs better than the state-of-the-art method.