Multi-Scale Aggregation and Scene Parsing for Crowd Counting

Abstract

With the acceleration of urbanization, the number of urban residents increases dramatically, and large-scale crowdy scenes become more common. There are many public security risks in these scenes. Therefore, it is of great significance to do crowd counting in the scenes. To solve these problems, a Multi-scale Aggregation and Scene Parsing (MASPNet) is proposed in the paper. To do the precision test and robustness test, MASPNet was employed on the ShanghaiTech dataset, UCF-QNRF dataset, and UCF Crowd Counting 50 (UCF_CC_50) dataset. The Experimental results demonstrate good performance on counting accuracy and robustness.