The realm of video surveillance has various methods to extract foreground. Background subtraction is one of the prime methods for automatic video analysis. The sensitivity of a meaningful event of interest is increased due to dampening effect of background changes and detection of false alarm. Hence the model can be strongly recommended to industries. This paper restricts the focus to one of the most common causes of dynamic background changes: that of swaying trees branches and illumination changes. To overcome the issue available in existing system, we propose a method called as twin background modeling. This method has dual models namely long and short term background models to increase exposure rate of foreground by using statistical method and also reduce false negative rate. This method has dimensional transformation from 2D to 1D which reduces computation time of the system and increases batch processing. The proposed method uses Manhattan distance to reduce execution time, increase detection rate and reduce error rate. The performance of the suggested approach is illustrated by using change detection dataset 2014 and is compared to other conventional approaches. © 2017, Springer Science+Business Media, LLC, part of Springer Nature.