Novel crowd synthesis approach and GLCM-derived descriptor for crowd behaviour detection. (24th December 2021)

• Record Type: Journal Article
• Authors: Hao, Yu; Liu, Ying; Fan, Jiulun; Gan, Yuquan
• Is Part Of: International journal of grid and utility computing. Volume 12:Number 5/6(2021)
• Page Start: 590
• Page End: 604
• Publication Date: 2021-12-24
• Subjects: crowd behaviours -- feature extraction -- image classification -- image texture analysis -- information entropy; Electronic data processing -- Distributed processing -- Periodicals; Electronic commerce -- Management -- Computer programs -- Periodicals; 004.605
• Languages: English
• ISSNs: 1741-847X

Abstract

Unlike an individual's behaviours, the dynamic nature of the crowd makes its behaviours difficult to be defined and classified. Therefore, techniques and approaches for crowd behaviour analysis usually encounter challenges different from an individual's behaviours. This paper aims to tackle several key issues in the procedure of crowd behaviour analysis. Firstly, a novel taxonomy is proposed to provide criteria for the explicit definition of different crowd behaviours. By adapting personal space, relative velocity and group force into conventional Social Force, a crowd behaviour synthesis approach is devised to provide visually realistic data for model training. Secondly, this paper introduces an effective entropy-based motion texture extraction algorithm, in order to accurately obtain the spatio-temporal motion information from the crowd. Furthermore, this paper proposes a novel visual descriptor based on Grey-Level Co-occurrence Matrix (GLCM) derived patterns to describe the visual essence of crowd anomalies. By applying proposed optimisations to the crowd behaviour analysis process, experiment results indicate the detection performance on panic dispersing and congestion is significantly increased.