Abstract

The fundamental working of background subtraction is to identify the moving region by taking pixel-wise difference of the current frame from the previous one. The proposed study presents the comparison and implementation of different background subtraction techniques i.e., frame-difference method, mixture of Gaussian model 2 (MOG2) and k-nearest neighbor (KNN) for background subtraction. For all the three techniques, prior to segmentation, background modeling and then features extraction steps are performed. It is investigated that on the same dataset, frame-difference technique outperforms both MOG2 and KNN and achieve accuracy of 89.98%, recall of 94.43% precision 79.55% and F1-score of 81.42%.