Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{203768,
        author = {Tarun Rathod and Mahesh Gunjal and Shreya Phakade and Gayatri Mandlik and Sakshi Pichad},
        title = {Facial Recognition-Based AI System Using CNN and DeepFace to Track Suspicious Activities and Identify Known Criminals in Real-Time Using CCTV Footage},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {164-172},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=203768},
        abstract = {In today’s world, surveillance cameras are every- where, yet most crimes caught on camera are noticed only after the fact—when someone spots something wrong. A person staring at six camera feeds all day simply cannot stay focused long enough to catch things as they happen. This paper describes a Crime Monitoring System (CMS) we designed and built to tackle exactly that problem. Our system watches live CCTV footage and, in real time, runs three separate deep learning models across parallel threads: YOLOv5 handles weapon detection, MobileNetV2 flags violent or abnormal behavior, and a Haar Cascade with LBPH pipeline handles face recognition. These three streams feed into a scoring formula we developed called IPETO—Identified Persons Excluding Trusted Officers— which calculates a live threat level and maps it onto one of four alert states. When the threat state changes, a WebSocket event fires inside 200 ms, and the operator’s dashboard updates automatically. The system regularly performed at 18–22 frames per second with all models enabled when we tried it on a laptop without a separate GPU. Face recognition scored 97%, weapon detection scored 93.2%, and violence recognition scored 94.1%. The complete pipeline, from camera input to browser-based alerts is implemented using a Flask-Socket IO web application. We used SQLite to store alert records.},
        keywords = {Crime Monitoring System, Deep Learning, YOLOv5, MobileNetV2, LBPH, CLAHE, Haar Cascade, Flask, WebSocket, SQLite, Real-Time Surveillance},
        month = {June},
        }