Detecting Video Anomalous Events with an Enhanced Abnormality Score

研究成果: 書籍/レポート タイプへの寄稿会議への寄与

抄録

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.

本文言語英語
ホスト出版物のタイトルPRICAI 2022
ホスト出版物のサブタイトルTrends in Artificial Intelligence - 19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022, Proceedings
編集者Sankalp Khanna, Jian Cao, Quan Bai, Guandong Xu
出版社Springer Science and Business Media Deutschland GmbH
ページ202-217
ページ数16
ISBN(印刷版)9783031208614
DOI
出版ステータス出版済み - 2022
イベント19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022 - Shangai, 中国
継続期間: 11月 10 202211月 13 2022

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
13629 LNCS
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

会議

会議19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022
国/地域中国
CityShangai
Period11/10/2211/13/22

!!!All Science Journal Classification (ASJC) codes

  • 理論的コンピュータサイエンス
  • コンピュータサイエンス一般

フィンガープリント

「Detecting Video Anomalous Events with an Enhanced Abnormality Score」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル