ProctorVision monitors entire exam halls using a single CCTV camera — detecting suspicious behavior through AI pose estimation, with zero cloud dependency.
Every feature designed around the constraints of physical examination environments
Separate thresholds for front, middle, and back rows — each student evaluated by their own bounding box height, not a single global threshold.
When ears are not visible, the system falls back to nose-to-shoulder geometry for head turn detection — works regardless of head covering.
Draw seat zones once using the ROI tool. System maps each detected person to their roll number and name automatically every session.
Automatically saves half-speed micro-clips and snapshots of violations with timestamps — ready for review or academic proceedings.
Detects when two adjacent students exhibit simultaneous suspicious behavior — flags potential coordination with a connecting alert.
One-click report generation per student — includes risk score, violation timeline, evidence photos, and micro-clips in a shareable HTML file.
From camera feed to evidence report in real time
RTSP stream from any IP or CCTV camera
YOLOv8-Pose extracts 17 skeleton keypoints per person
Pose geometry detects head turns, body lean, hand signals
Each person mapped to roll number via seating plan
Clips, photos, and HTML reports generated automatically
Runs on standard university hardware — no GPU or internet required
Windows 10 / 11 (64-bit)
Intel i5 / i7 (8th gen+)
8 GB RAM minimum
5 GB free disk space
Any RTSP / ONVIF IP camera
Local LAN only — no internet needed
Download the standalone executable — no Python or installation required
Download ProctorVision v1.1.0
Developed by Ahmed Muarij Siddiqui, Fatima Amin Siddiqui,
Nabiha Fatima & Abdul Raahim Sheikh
Supervised by Dr. Muhammad Wasim · Department of Software Engineering
UIT University · Batch 2022 · Final Year Project