Detecting Video Anomalous Events with an Enhanced Abnormality Score

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Detecting video anomalous events is vital for human monitoring. Anomalous events usually contain abnormal actions with exaggerated motion and little motion. We define the former and the latter as dynamic anomalies and static anomalies, respectively. We define the video data of events where a few persons perform diverse actions indoors as Indoor Event Data (IED). Many frame prediction approaches have succeeded in detecting dynamic anomalies. However, they are prone to overlooking static anomalies in IED. To solve this problem, we propose an Enhanced Abnormality Score (EAS), which is a combination of prediction, dynamic, appearance, and motion scores. To specifically target static anomalies, we calculate a score to evaluate the dynamic degrees of actions. We use an appearance score of a frame to detect static anomalies from appearance. This score is generated from a clustering-based distance of a pre-trained CNN feature. We also use a motion score based on flow reconstruction to balance the appearance score. We conduct extensive experiments on two datasets involving indoor human activities. Quantitative and qualitative experimental results show that our proposal achieves the best performance among its variants and the state-of-the-art methods.

Original languageEnglish
Title of host publicationPRICAI 2022
Subtitle of host publicationTrends in Artificial Intelligence - 19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022, Proceedings
EditorsSankalp Khanna, Jian Cao, Quan Bai, Guandong Xu
PublisherSpringer Science and Business Media Deutschland GmbH
Pages202-217
Number of pages16
ISBN (Print)9783031208614
DOIs
Publication statusPublished - 2022
Event19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022 - Shangai, China
Duration: Nov 10 2022Nov 13 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13629 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022
Country/TerritoryChina
CityShangai
Period11/10/2211/13/22

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Detecting Video Anomalous Events with an Enhanced Abnormality Score'. Together they form a unique fingerprint.

Cite this