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{205309,
        author = {Pushkar Mahajan and Arjun Deshmukh and Sumit Suryawanshi and Yash Ingale and Prof. Tanmayee Kute},
        title = {DRUSHTI-AI: A Smart Surveillance System},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {7458-7463},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=205309},
        abstract = {Academic examination environments continue to depend on human invigilation as the primary mechanism for malpractice prevention, a method constrained by operator fatigue, divided attention across large candidate populations, and inconsistent enforcement. This paper presents Drushti AI, an intelligent surveillance system that autonomously detects cheating behaviours in examination halls through continuous analysis of live video feeds. The system integrates YOLO11 Nano for real-time person detection, DeepSort for persistent multi-candidate identity tracking, and MediaPipe FaceMesh for granular facial landmark extraction. Behavioural violations are inferred via OpenCV-based head pose estimation computing yaw, pitch, and roll angles using the solvePnP algorithm alongside iris-based gaze deviation analysis and lip movement monitoring. Five malpractice categories are identified: lateral head deviation, sustained gaze redirection, lip movement, inter-candidate proximity, and prolonged face absence. A FastAPI backend orchestrates the detection pipeline with threaded multi-camera support and streams incident data to clients via WebSocket, persisting all detected events in a Supabase database. Operators interact through a Next.js web dashboard and a companion mobile application. Experimental evaluation on webcam- and CCTV-based video demonstrates reliable detection across all five categories with low false-positive rates at near-real-time processing speeds.},
        keywords = {Intelligent Surveillance, Examination Hall Monitoring, YOLO11, DeepSort, MediaPipe FaceMesh, Head Pose Estimation, Gaze Deviation Detection, Malpractice Detection, FastAPI, WebSocket, Supabase, Next.js},
        month = {June},
        }