An Efficient algorithm for background subtraction in videos

Abstract

Video processing is a substantially important branch of Image Processing and Computer Vision focused on extracting information from real scene videos. Among several other video processing techniques, Background Subtraction has attained great importance as a developing research area, during the past few years. It is a widely used technique particularly in surveillance videos, object tracking and detection, traffic or crowd monitoring etc. The goal of Background Subtraction is to segment the moving foreground part from the stationary background for a given scene, in order to make the post-processing tasks efficient and relatively easier. In this research, we propose a background subtraction technique that aims at progressively fitting a particular subspace for the background that is obtained from L1-Low rank matrix factorization (LRMF) using cyclic weighted median (CWM) and a certain distribution of mixture of Gaussian (MoG) for the foreground. Expectation maximization (EM) algorithm is applied to optimize the Gaussian mixture model (GMM). The effectiveness of the proposed method is augmented by using subsampling technique to execute on an average more than 250 frames per second while maintaining a good performance in accuracy. The performance of the proposed method is evaluated by comparing it with other state-of-the-art methods and it was concluded that the proposed method performs well in terms of F-measure and computational complexity.