Fast Moving Crowd Counting for Unconstrained Videos

Xinlei Wei,Junping Du,Dan Fan,JangMyung Lee
DOI: https://doi.org/10.1007/978-981-10-2338-5_46
2016-01-01
Abstract:It is important to estimate the number of people in fast moving crowd scenarios for the surveillance systems. Regression-based techniques achieved promising results for counting the number of people in crowded scenes. However, appearance features used in the most existing techniques are not able to mirror the motion state of crowds. The motion state of crowds is important to solve the counting problem of fast moving crowds, since it is decided by the amount of people. In this study, we propose a novel method to address this problem from three perspectives: (1) train a crowd counting estimation model suited to all of the crowded scenes; (2) combine motion states with multiappearance features for crowd counting; and (3) count fast moving crowds in unconstrained videos. These ideas are implemented in a fast object segment framework, which can segment fast moving crowds in the unconstrained videos. Extensive experiments validate the effectiveness of our proposed method.
What problem does this paper attempt to address?