Multi-source approach for crowd density estimation in still images
Sonu Lamba, Neeta Nain · 2017
Estimation of people density in intensely dense crowded scenes is very crucial due to perspective difference, few pixels per target, clutter and complex backgrounds etc. Most of the existing work is unable to handle the crowds of hundreds or thousands. At this level of density, one feature is not enough to estimate the total density of an image. We propose a hybrid model which relies on multiple source of information as Fourier analysis, Local binary pattern, Gray level dependence matrix (GLDM) features and Histogram of oriented gradient (HOG) for head detection to estimate the total count. Each of these features separately contribute in final total count estimation along with other statistical measures. Our approach is tested on hundred images of dense crowd annotated with 87K individuals. Experiential results validate the performance of our proposed approach by computing the total count with respect to ground truths.