郑州大学学报（理学版）2024，Vol.56Issue(1)：16-24,9.DOI:10.13705/j.issn.1671-6841.2022284
骨架引导的多模态视频异常行为检测方法
Skeleton-guided Multimodal Video Abnomaly Behavior Detection
摘要
Abstract
A multi-modal abnormal behavior detection algorithm based on the fusion of RGB video and skeleton sequences was proposed to solve the problem that the behavior of similar motion patterns was dif-ficult to distinguish by using only human skeleton features.In order to make full use of the advantages of each mode for abnormal behavior detection with similar behaviors,a new spatial embedding was used to strengthen the correspondence between RGB and skeletal poses,and temporal self-attention was used to extract the inter-frame relationship between the same nodes,which could effectively extract discriminative abnormal behavior features.In two large-scale public standard datasets,the results showed that the meth-od could achieve accurate detection of human abnormal behaviors compared with the good performance of spatiotemporal graph convolutional network detection algorithms when similar abnormal behaviors were in-distinguishable.关键词
视频异常行为检测/骨架/多模态融合/时空自注意力增强图卷积/空间嵌入Key words
video abnormal behavior detection/skeleton/multimodal fusion/spatiotemporal self-atten-tion augmented graph convolution/spatial embedding分类
信息技术与安全科学引用本文复制引用
付荣华,刘成明,刘合星,高宇飞,石磊..骨架引导的多模态视频异常行为检测方法[J].郑州大学学报（理学版）,2024,56(1):16-24,9.基金项目
国家重点研发计划项目(2018YFC0824402). （2018YFC0824402）
