Measuring Crowd Collectiveness via Global Motion Correlation

Ling Mei, Jianghuang Lai, Zeyu Chen, Xiaohua Xie; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

Abstract


Crowd collectiveness refers to the behavior consistency of crowd scenes, which reflects the degree of collective movements among massive individuals in crowd systems. The existing methods focus on measuring the discrepancy of motion direction among the individuals. However, few studies consider the magnitude discrepancy of velocity in a crowd and the collectiveness among different crowds, which can also affect the overall crowd collectiveness. In this paper, we propose a novel descriptor which combines intra-crowd collectiveness with inter-crowd collectiveness to solve the problem. For intra-crowd collectiveness, we introduce the energy spread process to identify the impacting factors of collectiveness, then measure the collectiveness of individuals within a crowd cluster by computing their similarities of magnitude and direction from the optical flow. For inter-crowd collectiveness, we assess the motion consistency among various crowd clusters generated from collective merging. Experimental results demonstrate that how the new collectiveness descriptor improves performance on three different crowd datasets, thus validating the superiority of the proposed descriptor.

Related Material


[pdf]
[bibtex]
@InProceedings{Mei_2019_ICCV,
author = {Mei, Ling and Lai, Jianghuang and Chen, Zeyu and Xie, Xiaohua},
title = {Measuring Crowd Collectiveness via Global Motion Correlation},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
month = {Oct},
year = {2019}
}