AI-Powered Real-Time Violence Detection in CCTV Feed

Abstract

We can observe that it is an ever-increasing problem of violence in the streets, in transport hubs, and other large areas of people, a tendency that has been developing over the past few years. We currently have predominantly conventional CCTV surveillance that is primarily reactive, and we rely heavily on human intervention to monitor numerous video feeds. In this manner, we find the method to be quite a burden on operators, resulting in response lag, and we are not getting early indications of violent action, which subsequently diminishes its value in prevention. In this regard, we present a practical, scalable, proactive AI-based system for real-time violence detection using CCTV video. A lightweight computer vision and deep learning framework is proposed for real-time violence detection using OpenCV for video capture and preprocessing. Mediapipe Pose extracts 2D/3D skeletal keypoints, which are processed by a TensorFlow-based BiLSTM model to learn spatiotemporal patterns of violent actions. Trained on benchmark datasets and real surveillance data, the system achieves 93.7% accuracy in classifying aggressive behaviors. A pose aggregation and sliding-window voting mechanism improves robustness in crowded and occluded scenes, reducing false positives below 4%. The system also triggers real-time SMS alerts via Twilio with contextual metadata for rapid response and intervention. There are extensive testing of edge devices and the cloud platform, with end-to-end alert latency below 1.3 seconds, including with high-resolution inputs, and it can be scaled linearly to multiple camera streams. The Framework encourages the deployment of privacy-aware systems by being privacy-conscious (using only anonymized pose representations) and by promoting on-device inference. The suggested solution demonstrates that it is possible to achieve accurate, low-latency, cost-effective violence detection with minimal hardware requirements and open-source software, which is why it can be used in schools, transportation hubs, commercial areas, and other public venues.