Estimation of Density Levels in the Holy Mosque from a Network of Cameras

Abstract

In this work we developed a systemAli, Yasir S.for estimating the density levelsZafar, Basimin the holy mosque of Makkah using video camerasSimsim, Mohammedinstalled in the mosque. This set-up relies on dividing the image into smaller segments and counting the number of people in each segment to infer the density. This algorithm used texture and SIFT interest point features to get an accurate count of the number of people at each segment using support vector regression. Having segments at different sizes helped to account for objects with different size in the image. In addition, the use of overlapping segment smooth the estimated density maps as each pixel receive a contribution from different patches. Our methodology has been tested extensively with different cameras during the Fasting season of 2015 with images from very crowded areas in the mosque.