Publication

Abstract : Estimation of the people count in a highly congested scene is a challenging task. It has an extensive range of applications in smart environments such as traffic monitoring, video surveillance, and automatic crowd management. It includes various challenges such as insufficient resolution, dynamic backgrounds, severe congestion, excessive overlaps, occlusions, and perspective changes. With the advancements in deep learning, crowd counting has been achieving excellent results in terms of accuracy and robustness. In this paper, we perform a brief review of perspective crowd counting and a study on recent perspective crowd counting methods namely S-DC Net, PCC Net, and SCAR. We have also identified a variant, which use both S-DC Net and Down Up Left Right (DULR) module and studied its performance on benchmark datasets. Few directions for further research are also presented.