An exam-monitoring prototype using YOLO, MediaPipe and OpenCV to analyze a student's head direction, flagging head turns in red for review and forward gaze in green.
An academic project to build a computer vision prototype for exam monitoring that analyzes camera frames to flag behavioral indicators that may be linked to cheating, chiefly head turning away from the forward direction. It was carried out by a team of students with technical assistance from Techno Enjaz, using Python and OpenCV, a YOLO model to locate the person in the scene, and MediaPipe to extract face and body landmarks.
In the implemented version, turning the head to the right or left, whether slightly or fully, is classified as a state that warrants review and a red frame is drawn around the student's face, while facing straight toward the screen is treated as normal and shown with a green frame. The prototype was tested with an ordinary camera in a controlled environment in terms of movement, lighting, and camera angle, not in real exam halls, and the project reports no numeric accuracy metrics of its own.
| Item | Details |
|---|---|
| Project type | Computer vision prototype for analyzing behavior during exams |
| Field | Education, artificial intelligence, computer vision, pose estimation |
| Project status | Prototype tested in a controlled environment with an ordinary camera |
| Techno Enjaz role | Technical support and assistance to the students during project development |
| Core technologies | Python, OpenCV, YOLO, MediaPipe, Visual Studio Code on Windows |
| Outputs | Real-time head-orientation classification, red frame for suspicious states and green frame for normal ones |
Exam monitoring traditionally relies on human invigilators, which faces clear difficulties: watching many students at once, inconsistent standards between invigilators, declining attention as fatigue builds over long exams, and high human and logistical costs in large exams. An invigilator may miss movements such as repeatedly turning toward a neighbor or shifting posture suspiciously.
The project proposes an assistive tool that analyzes students' movements from video frames using consistent, adjustable criteria, aiming to support the human invigilator rather than replace them and to provide indicators that can be reviewed later.
The project was carried out academically by the students, with technical assistance and support from Techno Enjaz during development. It is presented among the office's work as an example of its support for applied projects that combine object detection and pose estimation for behavior analysis.
The project documents an algorithm flowchart with the following steps:
| Tested state | System response |
|---|---|
| Slight or full turn to the left | Classified as suspicious, red frame around the face |
| Slight or full turn to the right | Classified as suspicious, red frame around the face |
| Face directed straight at the screen | Normal state, green frame around the face |
The project reports that the system responded immediately to changes in head direction and distinguished focus from deviation within the test scenario. This result is descriptive, however: the book does not state the number of cases tested or any project-specific quantitative metrics such as Precision, Recall, false-alarm rate, or frames per second. The accuracy figures that appear in the book (such as 95% and 94%) belong to earlier studies reviewed in the background chapter, not to this prototype.
According to the proposals set out in the project:
This kind of system processes images of students' faces and movements, so any real-world use requires informing students about the monitoring, defining its purpose, controlling access to recordings and how long they are kept, and keeping the final decision with a human invigilator.
If you are working on a computer-vision-based monitoring system and need technical help developing the idea or building a prototype, you can contact the Techno Enjaz office to discuss your project's requirements and the appropriate scope of support.
A computer vision prototype built by a team of students with technical assistance from Techno Enjaz that analyzes camera frames to flag behavioral indicators that may be linked to cheating, chiefly head turning. It is an assistive tool for the human invigilator, not a system that proves cheating.
The system loads a YOLO model to locate the person, captures camera frames, crops the person's image, converts it to RGB, and passes it to MediaPipe to extract face and body landmarks. It then analyzes the head's turning angle relative to the forward direction, drawing a red frame around the face for a head turn and a green frame when facing straight at the screen.
Python with OpenCV for frame processing and drawing, a YOLO model for person detection, and MediaPipe for pose estimation and landmark extraction, developed in Visual Studio Code on Windows.
In practical testing, the system classified slight or full head turns to the right or left as suspicious and facing straight ahead as normal, responding immediately to changes in head direction. The book does not report the number of test cases or metrics such as Precision or Recall, and the 95% and 94% figures it mentions belong to earlier studies, not this prototype.
The implementation only analyzes the head direction of one person in front of an ordinary camera in a controlled setting, without testing in crowded exam halls. A head turn can be innocent, so every flag needs human review, and instant alerts to invigilators and timestamped logs are stated as objectives but not shown as implemented.
Reach out to Techno Enjaz engineering team to discuss technical feasibility, prototype hardware, and development scope.
Discover more engineering prototypes and academic implementations
A virtual board for writing, drawing, picking colors and erasing with hand gestures in front of a camera, built in Python with MediaPipe hand tracking and OpenCV.
A prototype that detects people with YOLOv8 and tracks them with DeepSORT in defined work areas, counts presence time in seconds, and exports records to Excel.
A face-recognition access control prototype using OpenCV and face-recognition that logs authorized entries, saves images of unknown faces, and sounds an alarm.
