International Journal of Scientific Research in Network Security and Communication

Real-time background/foreground separation of a video is necessary for detecting an object, identifying, tracking vehicle, as well as recognizing objects. Several algorithms were already found for background initialization and foreground detection. Recent examinations have presented a robust method, Dynamic Mode Decomposition (DMD) for separating video frames into a background (low-rank) model and foreground (sparse) segments. The full video stream first converted to frames and applied DMD on each frame for the separation of background/foreground objects. Since the method uses full video frames it had more computational complexity and difficult to analyze. For a better solution, this paper shows that the large continuous video stream first converted to frames and each frame breaks into segments and then DMD applied on the segment where the moving object or foreground is obtained. The issue using DMD is the burden of working with large information and now it can be easy work using the segmented video frames. The strength of DMD is demonstrated using a publicly available Scene Background Initialisation (SBI) dataset. The objective of this work is to obtain a background/foreground model from a video sequence where the background is filled with a number of foreground objects with less complexity. Finally compared the accuracy parameters between different background/foreground separation methods with DMD and shows the performance of DMD_Segmented is much better.