A Real-Time Intelligent Video Surveillance Framework for Automated Violence Detection Using Advanced Deep Learning Architectures

2025 · Video Understanding / Public Safety AI

A real-time surveillance framework for automated violence detection using deep learning architectures (CNN-LSTM & Attention) tuned for fast and reliable scene-level threat recognition.

Research Focus

Real-time violence detectionDeep spatiotemporal modelingCNN-LSTM architecturesPose estimation

Publication Type

Research Paper

Area

Computer Vision

Summary

This paper proposes a real-time intelligent monitoring pipeline that classifies violent behavior directly from video streams. The architecture combines advanced deep learning components (pose-based spatiotemporal features, CNN-LSTM, Attention mechanisms) for robust temporal and spatial understanding, then triggers automated alerts for rapid intervention. The framework achieved an accuracy of 95.6% and is designed for practical deployment in safety-critical public and private environments.

Open Paper ↗Download PDF

Next Research >