INTELLIGENT SURVEILLANCE SYSTEM FOR REAL-TIME CRIME DETECTION AND BEHAVIOUR ANALYSIS

Abstract

Conventional surveillance depends heavily on continuous human observation of camera feeds, which can lead to fatigue, delayed threat recognition, and inconsistent response. This paper presents an intelligent surveillance framework for real-time crime detection and behavior analysis using computer vision and deep learning. The proposed approach processes live camera input through OpenCV, detects relevant objects using YOLO, extracts visual patterns using a convolutional neural network (CNN), and analyses activity sequences using Long Short-Term Memory (LSTM). Detected events are assigned threat levels and can trigger automated notifications containing incident information. The architecture also considers multi-camera monitoring, edge processing and cloud-based storage to improve responsiveness, scalability and incident management. An estimated comparison of the techniques reported in the project indicates that CNN provides the strongest overall classification performance, while YOLO is particularly suitable for real-time object detection and LSTM contributes to temporal behavior analysis. The proposed framework reduces dependence on manual monitoring and provides a practical foundation for faster, more consistent and proactive surveillance.