A Computer-Vision-Augmented Web-Based Online Examination Proctoring System with Privacy-Preserving Behavioural Monitoring: Design, Implementation and Evaluation

The rapid expansion of digital education and internet-based learning has created an urgent, unresolved need for secure, reliable, and intelligent online examination systems. Traditional physical invigilation cannot scale to geographically distributed student populations or hybrid institutional modes. This paper presents an Online Examination Proctoring Systema full-stack Django web application that integrates computer-vision-based behavioural monitoring with complete examination management, automated evaluation, and a cryptographically anchored audit trail. The system implements four complementary layers: a responsive browser Presentation Layer using Bootstrap 5 and JavaScript; a Django MVT Application Layer handling authentication, question bank management, examination delivery, autograding, and reporting; a Monitoring Layer combining server-side OpenCV Haar-cascade face detection (face-absent / multiple-face conditions) with client-side JavaScript event listeners (tab-switch, focus-loss, page-unload); and a SQLite Data Layer persisting seven core models -User, UserProfile, Examination, Question, ExamSession, Response, and ProctorLog. A privacy-preserving design principle is central: webcam frames are processed on the institution's own server without streaming to any third-party cloud service, and no client-side software installation beyond a standard browser is required. Multi-format objective questions (MCQ, true/false) are auto-graded with instant result generation;short-answer items are queued for administrator review.System validation through unit, integration, and useracceptance testing confirms detection accuracy of 94.2% for face-absent events, 91.5% for multiple-face events, 98.7% for tab-switch capture, and 100% auto-grading accuracy against ground-truth answer keys.The system is fully deployable on commodity hardware without proprietary software dependencies, providing an accessible, privacyrespecting, and open-source alternative to expensive commercial proctoring platforms.