Development of a Real-Time CCTV-Based Violence Detection System

In modern urban life, ensuring public safety presents a significant challenge due to the complexity of city infrastructure and the increasing demand for real-time surveillance solutions. This research involves the development of a Smart City CCTV Monitoring System that integrates deep learning and computer vision to automatically detect and classify violent incidents and weapons. The system uses Convolutional Neural Networks (CNNs) to analyze video frames extracted from continuous CCTV streams at regular intervals. For object and action recognition, it employs advanced models such as YOLO for object detection and Mediapipe for pose estimation. Every 5 seconds, the system extracts video frames, treating them as images that are processed by a trained CNN-LSTM model for real-time predictions. The model was trained on a combination of public datasets, including the Crow Violence Dataset, which contains over X labeled video clips of both violent and non-violent scenarios. During performance evaluation, the system achieved an accuracy of 94.7%, precision of 93.5%, recall of 95.2%, and an F1-score of 94.3%. These metrics demonstrate both high sensitivity for detecting violent activities and a low false positive rate. By automatically generating real-time alerts for law enforcement and emergency services, the system aims to significantly reduce response times and prevent potential harm. This comprehensive approach enhances the reliability and efficiency of surveillance measures, contributing to the vision of safer and smarter cities.