Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ComplexVAD: Detecting Interaction Anomalies in Video

Furkan Mumcu, Michael J. Jones, Yasin Yilmaz, Anoop Cherian

TL;DR

A novel method to detect complex anomalies via modeling the interactions between objects using a scene graph with spatio-temporal attributes is proposed and it is demonstrated that this new method outperforms existing works.

Abstract

Existing video anomaly detection datasets are inadequate for representing complex anomalies that occur due to the interactions between objects. The absence of complex anomalies in previous video anomaly detection datasets affects research by shifting the focus onto simple anomalies. To address this problem, we introduce a new large-scale dataset: ComplexVAD. In addition, we propose a novel method to detect complex anomalies via modeling the interactions between objects using a scene graph with spatio-temporal attributes. With our proposed method and two other state-of-the-art video anomaly detection methods, we obtain baseline scores on ComplexVAD and demonstrate that our new method outperforms existing works.

ComplexVAD: Detecting Interaction Anomalies in Video

TL;DR

A novel method to detect complex anomalies via modeling the interactions between objects using a scene graph with spatio-temporal attributes is proposed and it is demonstrated that this new method outperforms existing works.

Abstract

Existing video anomaly detection datasets are inadequate for representing complex anomalies that occur due to the interactions between objects. The absence of complex anomalies in previous video anomaly detection datasets affects research by shifting the focus onto simple anomalies. To address this problem, we introduce a new large-scale dataset: ComplexVAD. In addition, we propose a novel method to detect complex anomalies via modeling the interactions between objects using a scene graph with spatio-temporal attributes. With our proposed method and two other state-of-the-art video anomaly detection methods, we obtain baseline scores on ComplexVAD and demonstrate that our new method outperforms existing works.
Paper Structure (20 sections, 11 equations, 12 figures, 2 tables)

This paper contains 20 sections, 11 equations, 12 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. ComplexVAD
  4. Detecting Complex Anomalies
  5. Frame to graph
  6. Model building from nominal video
  7. Complex anomaly detection in test video
  8. Experiments
  9. Experimental settings and evaluation criteria
  10. Results
  11. Future Work and Discussions
  12. Conclusion
  13. More details on ComplexVAD
  14. Dataset Details
  15. Anomaly Types
  16. ...and 5 more sections

Figures (12)

  • Figure 1: Nominal frame samples from the ComplexVAD dataset.
  • Figure 2: Samples of complex anomaly from the ComplexVAD dataset. In all samples, objects are nominal, but the interactions are anomalous. (Top) A skateboard moving alone violates the expectation of a nominal interaction between a person and skateboard. (Middle) Person carries an object and then leaves it on the ground. (Bottom) Three people walk together and then suddenly start to run in different directions.
  • Figure 3: The pipeline of our method for frame to graph generation with the help of an object detector, object tracker and pose estimator.
  • Figure 4: A person who drops a bag on the street is detected as an anomaly with our method. The detection starts with the action of "drop". After the object interaction ends, the dropped object continues to be detected as an anomaly. Ground truth labels and detection boxes are represented with green and red colors, respectively.
  • Figure 5: Detected interaction anomalies with our method. (Top) A dog without a walker. (Middle) Car picks up a passenger on crosswalk. (Bottom) Bicycle stops briefly in the middle of the road. Ground truth labels and detection boxes are represented with green and red colors, respectively.
  • ...and 7 more figures