A foreground content centric efficient surveillance video coding with adaptive GoP model and singular value decomposition

Abstract Video surveillance is related to our current social scenario, where this type of service is rigorously increasing day by day. The video surveillance system generates a massive volume of data, and the storage requirement for this data is challenging. In this paper, we have proposed a novel video compression algorithm for surveillance video. We aim to compress the surveillance video data. A static camera fixed at a surveillance site captures surveillance video. The proposed method breaks the surveillance video data into multiple GoP, where each GoP corresponds to the number of objects present in the frame. We have preprocessed the video data by background subtraction to get residual video data that helps to achieve significant video compression. Residual video data is used for motion estimation and compensation to make the compression process more effective. By applying blockbased SVD on both the I-frame and the remaining frames within the GoP, we can protect the salient characteristics of the video data while achieving more efficient compression. We have tested our algorithm on the standard data set and some real data set collected by raspberry pi camera. The suggested algorithm is efficient in surveillance video data compression, preserving the acceptable visual quality of the compressed video.