Crowd counting using adaptive segmentation in a congregation

Abstract

A novel technique of segmenting people for counting using adaptive thresholding has been proposed in this paper. The technique uses adaptive thresholding to detect and extract objects of interest and is based on the idea of counting people through a slit window. The technique works well for highly occluded crowded scenes in Masjid-e-Haram during Hajj and Umrah congregation. It is a very diverse problem and presents variety of challenges to solve. The most challenging problem is to differentiate between ever-changing background and foreground. The density levels of crowd are also variable from single persons to dense crowd. In this paper, idea and practical application of the proposed method under real conditions has been described. A series of experiments were conducted on real videos of the event of variable densities and lengths. Finally, experiments proved that, proposed method is applicable in real time scenario with high accuracy.