Edge Guard AI: An Intelligent Real-Time CCTV Surveillance and Smart Security Analytics Framework Using Local Edge Processing

Conventional CCTV systems record continuous video, but identifying important incidents often depends on a person watching feeds or reviewing recordings after an event. In this work, Edge Guard AI, a proposed intelligent CCTV analytics framework, is presented for adding automated, locally executed video analysis to existing camera infrastructure. The system receives live video from CCTV cameras, IP cameras, webcams, or DVR/NVR streams and processes frames on a desktop or laptop computer, reducing the need to send complete video streams to cloud servers and addressing bandwidth, latency, cost, and privacy concerns. YOLO detects persons, weapons, and other predefined security-related objects; MTCNN detects and extracts faces; and FaceNet generates facial embeddings for comparison with registered records. Detected persons and events are classified as familiar, unknown, missing-person record matches, or law-enforcement watchlist record matches, with additional trained models supporting masked-face and suspicious-activity detection. Events are assigned Normal, Warning, or Danger alert levels, and significant events trigger image capture, short event-specific video recording, timestamping, and notification through a connected application, with a searchable event history. Role-based Admin, User, and Police modules manage records, alerts, and evidence. The framework is at the proposal and design stage; the capabilities described are objectives to be implemented and evaluated, not completed test results