PERTANIKA JOURNAL OF SCIENCE AND TECHNOLOGY

The increasing reliance on surveillance systems underscores the critical need for efficient and automated solutions to detect rare events within residential environments. This study introduces a robust anomaly detection framework for near real-time analysis of recorded CCTV footage. The proposed system addresses law enforcement's significant challenge: the manual, error-prone, and time-consuming review of extensive surveillance footage. The system integrates advanced techniques such as data stream management, human subject detection, and anomaly detection. A pre-trained Mask Region-Based Convolutional Neural Network (Mask R-CNN) ensures precise human detection, even in complex scenes. At the same time, an enhanced unsupervised Isolation Forest algorithm identifies rare or anomalous events by distinguishing them from routine activities in the video data. Results demonstrate that the framework significantly improves the speed and accuracy of crime analysis, offering a reliable, scalable, and automated solution with minimal human intervention. The system empowers law enforcement agencies and security professionals by streamlining detection, facilitating proactive responses, and thorough investigations. Its adaptability across diverse residential settings further strengthens its utility for various security and surveillance applications, ultimately contributing to enhanced safety and security measures.