Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Clustering Aided Weakly Supervised Training to Detect Anomalous Events in Surveillance Videos

Muhammad Zaigham Zaheer, Arif Mahmood, Marcella Astrid, Seung-Ik Lee

TL;DR

A weakly supervised anomaly detection system that has multiple contributions including a random batch selection mechanism to reduce interbatch correlation and a normalcy suppression block (NSB) which learns to minimize anomaly scores over normal regions of a video by utilizing the overall information available in a training batch is proposed.

Abstract

Formulating learning systems for the detection of real-world anomalous events using only video-level labels is a challenging task mainly due to the presence of noisy labels as well as the rare occurrence of anomalous events in the training data. We propose a weakly supervised anomaly detection system which has multiple contributions including a random batch selection mechanism to reduce inter-batch correlation and a normalcy suppression block which learns to minimize anomaly scores over normal regions of a video by utilizing the overall information available in a training batch. In addition, a clustering loss block is proposed to mitigate the label noise and to improve the representation learning for the anomalous and normal regions. This block encourages the backbone network to produce two distinct feature clusters representing normal and anomalous events. Extensive analysis of the proposed approach is provided using three popular anomaly detection datasets including UCF-Crime, ShanghaiTech, and UCSD Ped2. The experiments demonstrate a superior anomaly detection capability of our approach.

Clustering Aided Weakly Supervised Training to Detect Anomalous Events in Surveillance Videos

TL;DR

A weakly supervised anomaly detection system that has multiple contributions including a random batch selection mechanism to reduce interbatch correlation and a normalcy suppression block (NSB) which learns to minimize anomaly scores over normal regions of a video by utilizing the overall information available in a training batch is proposed.

Abstract

Formulating learning systems for the detection of real-world anomalous events using only video-level labels is a challenging task mainly due to the presence of noisy labels as well as the rare occurrence of anomalous events in the training data. We propose a weakly supervised anomaly detection system which has multiple contributions including a random batch selection mechanism to reduce inter-batch correlation and a normalcy suppression block which learns to minimize anomaly scores over normal regions of a video by utilizing the overall information available in a training batch. In addition, a clustering loss block is proposed to mitigate the label noise and to improve the representation learning for the anomalous and normal regions. This block encourages the backbone network to produce two distinct feature clusters representing normal and anomalous events. Extensive analysis of the proposed approach is provided using three popular anomaly detection datasets including UCF-Crime, ShanghaiTech, and UCSD Ped2. The experiments demonstrate a superior anomaly detection capability of our approach.
Paper Structure (19 sections, 10 equations, 4 figures)

This paper contains 19 sections, 10 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Anomaly Detection as One Class Classification (OCC)
  4. Anomaly Detection as Weakly Supervised Learning
  5. Clustering as Supervision
  6. Normalcy Suppression
  7. Proposed CLAWS Net+ Architecture
  8. Training Data Organization Block
  9. Video Segment Formation
  10. Feature Extraction
  11. Batch Formation
  12. Random Batch Selector
  13. Backbone Network Block
  14. Normalcy Suppression Block
  15. Clustering Loss Block
  16. ...and 4 more sections

Figures (4)

  • Figure 1: CLAWS Net+: The proposed weakly supervised anomaly detection framework using video-level labels. (a) Each input video is divided into equal length segments. (b) & (c) Using each video segment, a feature vector is extracted. (d) Feature vectors are arranged into batches maintaining temporal order. (e) For training, batches are randomly selected. (f) Backbone network block consists of FC Module-1 & 2. (g) & (h) Normalcy suppression block consists of normalcy suppression modules, NSM-1 & 2. (i) Clustering loss block in which a loss is computed using two clusters created in an unsupervised fashion.
  • Figure 2: Clustering Loss Block (CLB): intermediate feature representations from a complete video are divided into two clusters in an unsupervised fashion to compute the clustering loss. This loss helps the backbone network to learn more discriminative feature representations for normal and anomalous events.
  • Figure 3: Clustering loss serves two purposes. First, it encourages the backbone network to produce closer clusters in case of a normal video and distant clusters in case of an anomalous video. Second, it encourages the network to produce more compact clusters.
  • Figure 4: ROC curve comparison of the proposed approach with existing SOTA on UCF-Crime Dataset.