A Novel Methodology for Detecting Abnormal Activities in Video Surveillance using Weakly Supervised Learning

View at publisher Summarise with VIRAT
Authors: NAZREEN BANU A, Pavithiraa S, Nithiksha N, Nithikaa S J
Cited by: 0
References: 15

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Video surveillance systems have become integral parts of the modern safety and security infrastructure, although the traditional methods of monitoring systems have a difficult time identifying anomalies reliably in a complex and dynamic environment. To solve these challenges, this paper proposes an intelligent and multi-layer video anomaly detection architecture that integrates multimodal sensing, deep learning, and graphbased reasoning. The system starts with a multimodal Data Acquisition Layer to acquire RGB, thermal and infrared video streams, followed by a Pre-Processing Layer which includes stabilization, denoising, background subtraction, synchronization, sampling and dataset balancing. Spatialtemporal features are extracted using CNN based encoders, which are further converted into graph representations for relational modelling. A Weakly Supervised Graph Attention Network (GAT) is used in the Decision Support Layer to emphasize the semantically important nodes and deduce the abnormal events via attention-weighted message passing. Anomaly scores are obtained at the segment level and system performance is analyzed by using ROC and AUC metrics. Experimental results show that the proposed model of GAT has the best performance, i.e., 96.3% accuracy, 95.2% accuracy, and significant improvements after optimization. The Actuation Layer offers real-time visualization and automated alerts for real-time response. The results demonstrate the benefit of graph-based weak supervision for large-scale surveillance and spur advancements in lightweight and edge-deployable models for real-time environments.

Fields of Study & Subjects

Anomaly Detection Techniques and Applications Video Surveillance and Tracking Methods Human Pose and Action Recognition

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1800–1805
Publication date
21 January 2026
Language
EN