Real-time Violence Detection Using Deep Learning

Abstract

Violence Detection In Surveillance Videos Has Become An Essential Requirement For Modern Safety Systems Due To Rising Security Concerns Across Public, Private, And Institutional Environments. Traditional CCTV Monitoring Systems Rely Heavily On Manual Observation, Which Is Prone To Human Error, Delayed Response, And Inefficiency During High-risk Situations. Recent Advancements In Deep Learning And Computer Vision Have Enabled Intelligent Surveillance Systems Capable Of Identifying Violent Activities Automatically And In Real Time. This Paper Presents A Deep-learning-based Framework For Real-time Violence Detection Using MobileNetV2 And OpenCV. Transfer Learning, Data Augmentation, And Fine- Tuning Techniques Were Used To Enhance Model Accuracy And Address Data Imbalance. A Prediction-smoothing Algorithm Based On Moving Average Improves Detection Stability By Minimizing Frame-level Noise. The System Triggers An Audio Alert During Violence Events To Enable Immediate Intervention. Experimental Results Demonstrate That The Proposed Model Achieves High Accuracy, Stable Real-time Performance, And Successful Detection In Live Video Streams. This Solution Can Be Deployed In Institutions, Smart Cities, Public Spaces, And Home Monitoring Applications.

Keywords