Time Driven Video Summarization using GMM

In this paper, we propose a method to browse the activities present in the longer videos for the user defined time. Browsing of activities is important for variety of applications and consumes large amount of viewing time for longer videos. The aim is to generate a summary of the video by retaining salient activities in a given time. We propose a method for selection of salient activities using motion of feature points as a key parameter, where the saliency of a frame depends on total motion and specified time for summarization. The motion information in a video is modeled as a Gaussian mixture model (GMM), to estimate the key motion frames in the video. The salient frames are detected depending upon the motion strength of the keyframe and user specified time, which contributes for the summarization keeping the chronology of activities. The proposed method finds applications in summarization of surveillance videos, movies, TV serials etc. We demonstrate the proposed method on different types of videos and achieve comparable results with stroboscopic approach and also maintain the chronology with an average retention ratio of 95%.