ComplexVAD: Detecting Interaction Anomalies in Video

  • 2025-01-16 18:35:45
  • Furkan Mumcu, Michael J. Jones, Yasin Yilmaz, Anoop Cherian
  • 0

Abstract

Existing video anomaly detection datasets are inadequate for representingcomplex anomalies that occur due to the interactions between objects. Theabsence of complex anomalies in previous video anomaly detection datasetsaffects research by shifting the focus onto simple anomalies. To address thisproblem, we introduce a new large-scale dataset: ComplexVAD. In addition, wepropose a novel method to detect complex anomalies via modeling theinteractions between objects using a scene graph with spatio-temporalattributes. With our proposed method and two other state-of-the-art videoanomaly detection methods, we obtain baseline scores on ComplexVAD anddemonstrate that our new method outperforms existing works.

 

Quick Read (beta)

loading the full paper ...