Background Subtraction based on Geometric - K mean Algorithm

Abstract

Background Subtraction is widely used for detection of moving object, traffic management, and video surveillance even when the environmental condition is not favourable such as illumination problem, weather condition and fast moving object. Background subtraction methodology is extensively utilized in the detection of moving entity which is captured through a camera. Foundation for this methodology to identify the moving entity by observing the variation among the input frame and reference frame is described as background image. Fundamentally, background image is an illustration of section of images with no moving entity and that should be consistently modified to adjust with the changing illumination and geometric adjustments. Further composite prototypes were stretched the perception of background subtraction within the accurate significance. In this research, background modelling of Geometric Mean (GM) based lognormal distribution of each pixel is considered, followed by K-mean clustering algorithm is used to separate background from foreground. Finally to enhance the result weighted median filter is used. The proposed algorithm has been tested upon different data sets and the results shows better precision as compared to its ground truth by calculating sensitivity, specificity and accuracy.