Violence Detection in Video Using Advanced CV Algorithms
P. AKSHITH · INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 2025
This Investigation introduces an innovative dual-model framework for detecting violent content in video by synergistically combining YOLO (You Only Look Once) and RCNN (Region-based Convolutional Neural Networks) architectures. Our approach addresses the critical challenge of automated violence detection for applications in public safety monitoring, content moderation platforms, and surveillance systems. The proposed methodology leverages YOLO's computational efficiency with RCNN's precision through a novel weighted fusion algorithm that automatically balances their complementary strengths. Extensive evaluations across diverse datasets demonstrate that our hybrid system achieves 92.4% accuracy, surpassing single-model implementations while maintaining processing speeds suitable for practical deployment. The framework was rigorously tested on Hockey Fights, Movie Scenes, and Surveillance datasets, consistently demonstrating robust performance across varying environmental conditions, camera angles, and violence types. Key Words: Computer Vision, YOLO, CNN, RCNN.