A Sparse Pooling Adversarial Learning Framework for Anomaly Event Detection

Detecting abnormal event in video is essential for maintaining safety in modern communities. However, due to factors of complex background, large changes in scale, and the randomness of abnormal events, causing abnormal event detection poses significant challenges. To address the issue, we propose a...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Advances in Electrical and Computer Engineering Jg. 25; H. 2; S. 49 - 58
Hauptverfasser: ZHANG, M., HU, H., LI, Z.
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Suceava Stefan cel Mare University of Suceava 01.06.2025
Schlagworte:
Ablation
Analysis
anomaly detection
Artificial intelligence
Automation
Behavior
Comparative analysis
Datasets
Deep learning
Effectiveness
encoder-decoder
Frames (data processing)
generative adversarial networks
Image processing
Learning
Machine learning
Methods
Neural networks
self-attention
smart community
Surveillance
China
ISSN:1582-7445, 1844-7600
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Detecting abnormal event in video is essential for maintaining safety in modern communities. However, due to factors of complex background, large changes in scale, and the randomness of abnormal events, causing abnormal event detection poses significant challenges. To address the issue, we propose an effective sparse pooling adversarial learning framework (SPLF) for anomaly event detection, which integrates self-attention and pyramid features into a unified architecture. Specifically, the network takes video frames as input and employs an efficient U-Net to predict unknown frames. Meanwhile, self-attention mechanism and pyramid pooling features are combined to focus on salient areas and capture moving objects with varying scales. In addition, to evaluate the scores of abnormal events, a multi-scale error pyramid is introduced to improve the accuracy and robustness of the proposed SPLF. The comparison test is conducted on three publicly datasets: Ped2, Avenue, ShanghaiTech and a community scenario dataset. The frame-level AUC (area under curve) achieves 97.5%, 89.2%, 75.1% and 70.2% respectively, reaching a high level. Ablation tests further validate the effectiveness of self-attention mechanism and multi-scale pyramid pooling features. The test results demonstrate that the proposed method can effectively learn action patterns and accurately detect abnormal events in community scenarios. Index Terms--smart community, anomaly detection, encoder-decoder, generative adversarial networks, self-attention.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1582-7445
1844-7600
DOI:10.4316/AECE.2025.02006