/ README.md
README.md
1 # cheatshield 2 # AI-Based Online Cheating Prevention System in Online Exams 3 4  5 *Ensuring academic integrity in online exams through AI-powered proctoring.* 6 7 --- 8 9 ## 📝 Overview 10 11 The **AI-Based Online Cheating Prevention System** is an intelligent proctoring solution designed to maintain fairness and integrity during online examinations. With the rise of remote learning, ensuring secure virtual assessments has become a critical challenge. This system leverages advanced artificial intelligence techniques to monitor candidates in real time, detect cheating behaviors, and enforce exam rules, reducing the need for human invigilators while fostering trust in digital education. 12 13 The system integrates multiple detection modules—**object detection (YOLOv5)**, **face detection (MTCNN)**, and **audio detection (YAMNet)**—alongside system control mechanisms to create a comprehensive invigilation framework. It monitors for prohibited items (e.g., mobile phones, books), ensures the candidate’s presence, detects suspicious sounds (e.g., speech, whispering), and enforces rules like fullscreen mode and shortcut restrictions. A user-friendly interface built with **PyQt5** displays live webcam feeds, violation counts, and warnings, empowering proctors to oversee exams effectively. 14 15 --- 16 17 ## 🚀 Features 18 19 - **Object Detection**: Identifies prohibited items like mobile phones and books using YOLOv5 with a custom-trained model (`best.pt`). 20 - **Face Detection**: Ensures the candidate’s presence and flags multiple faces using MTCNN, preventing impersonation. 21 - **Audio Detection**: Detects suspicious sounds like speech or whispering using YAMNet, with a lowered threshold for improved sensitivity. 22 - **System Control**: Enforces exam rules using `QWebEngineView`, maintaining fullscreen mode and disabling shortcuts (e.g., Alt+Tab, F11). 23 - **Real-Time Monitoring**: Displays live webcam feeds, violation counts, and warnings in a PyQt5-based interface. 24 - **Violation Handling**: Implements a 15-second cooldown between violations and ends the test after 10 violations. 25 - **Resource Optimization**: Reduces webcam update rate to 10 FPS, lowers resolution to 160x120, and adjusts detection frequencies for smooth performance. 26 - **Scalability**: Designed to handle multiple candidates, making it suitable for large-scale institutional assessments. 27 - **Transparency**: Logs violations with timestamps for post-exam review and dispute resolution. 28 29 --- 30 31 ## 🛠️ Technologies Used 32 33 - **Python 3.x**: Core programming language for the system. 34 - **YOLOv5**: For object detection to identify prohibited items like mobile phones and books. 35 - **MTCNN**: For face detection to ensure candidate presence and prevent impersonation. 36 - **YAMNet**: For audio detection to flag suspicious sounds like speech or whispering. 37 - **PyQt5**: For building the user interface, displaying webcam feeds, and showing warnings. 38 - **QWebEngineView**: For enforcing exam rules like fullscreen mode and shortcut disabling. 39 - **TensorFlow Hub**: For loading pre-trained YAMNet model for audio detection. 40 - **OpenCV**: For webcam frame processing and display. 41 - **NumPy**: For numerical computations and array manipulations. 42 - **SoundDevice**: For audio capture and processing. 43 44 --- 45 46 ## 📂 Project Structure 47 AI-Based-Online-Cheating-Prevention-System 48 ``` 49 │ 50 ├── models/ 51 │ ├──best.pt # Custom-trained YOLOv5 model for object detection 52 ├── modules/ # Source code directory 53 │ ├── object_detection.py # Module for YOLOv5 object detection 54 │ ├── face_detection.py # Module for MTCNN face detection 55 │ ├── audio_detection.py # Module for YAMNet audio detection 56 │ ├── system_control.py # Module for enforcing exam rules (fullscreen, shortcuts) 57 ├── index.html # Frontend UI that appears on the external website 58 ├── styles.css # Stylesheet for the UI 59 ├── main.py # Main script to run the proctoring system 60 ├── requirements.txt # List of required Python packages 61 └── README.md # Project documentation 62 ``` 63 64 --- 65 66 ## ⚙️ Installation and Setup 67 68 ### Prerequisites 69 - Python 3.8 or higher 70 - A webcam and microphone 71 - Internet connection (for initial package installation) 72 73 ### Steps to run this project 74 1. **Clone the Repository** 75 ```bash 76 git clone https://github.com/vishal017/cheatshield.git 77 cd AI-Based-Online-Cheating-Prevention-System 78 ``` 79 2. **Create a Virtual Environment (Optional but Recommended)** 80 ``` 81 python -m venv venv 82 ``` 83 ***For Windows*** 84 ``` 85 venv\Scripts\activate 86 ``` 87 ***For Mac*** 88 ``` 89 source venv/bin/activate 90 ``` 91 3. **Install Dependencies** 92 ``` 93 pip install -r requirements.txt 94 ``` 95 4. **Run the project** 96 ``` 97 python main.py 98 ``` 99 100 ## 🖥️ Usage 101 102 1. **Launch the Application**: Run `main.py` to start the proctoring system. 103 2. **Exam Setup**: The system loads the exam website in a `QWebEngineView` window, enforcing fullscreen mode and disabling shortcuts. 104 3. **Monitoring**: The PyQt5 interface displays the webcam feed, violation count, and warnings in real-time. 105 4. **Violation Handling**: 106 - Issues warnings for detected violations (e.g., mobile phone detected, multiple faces, speech). 107 - A **15-second cooldown** prevents rapid triggers. 108 - The test **ends after 10 violations**. 109 5. **End Test**: 110 - The test can be ended **manually via a password-protected dialog** or **automatically after 10 violations**. 111 - Violations are logged in the `logs/` directory with timestamps. 112 113 --- 114 115 ## 📊 Performance and Outcomes 116 117 - **Enhanced Supervision**: MTCNN ensures continuous face detection with high reliability, flagging impersonation attempts. 118 - **Precise Object Detection**: YOLOv5 detects prohibited items accurately with confidence thresholds: 119 - **Mobile Phones** → `0.4` 120 - **Books** → `0.2` 121 - **Effective Audio Detection**: YAMNet detects speech with a **0.3 threshold**, optimized with a **5-second detection interval**. 122 - **System Control**: `QWebEngineView` enforces exam rules, preventing unauthorized access. 123 - **Resource Efficiency**: Optimized to run smoothly even during YouTube playback with: 124 - **10 FPS webcam update rate** 125 - **160x120 resolution** 126 - **Reduced detection frequencies** 127 - **Scalability**: Can handle **multiple candidates**, making it suitable for large-scale exams. 128 129 --- 130 131 ## 🔍 Challenges and Solutions 132 133 - **High Resource Usage** 134 - **Solution**: Reduced webcam FPS (`10`), lowered resolution (`160x120`), and optimized detection frequencies (`every 8th frame` for object/face detection, `every 5 seconds` for audio). 135 136 - **Misclassification in Object Detection** 137 - **Solution**: Improved by setting confidence thresholds (`0.4` for mobile phones, `0.2` for books) and using a **custom-trained YOLOv5 model**. 138 139 - **White Box Issue in UI** 140 - **Solution**: Made the warning display (`QWebEngineView`) **transparent** in the PyQt5 interface. 141 142 - **Noisy Audio Environments** 143 - **Solution**: Lowered **YAMNet speech detection threshold** (`0.3`) and added **debug prints** to verify audio capture. 144 145 --- 146 147 ## 🌟 Future Enhancements 148 149 - **Lighter Models**: 150 - Replace YOLOv5 with a **smaller variant (`yolov5s`)**. 151 - Use a **lighter face detection model** instead of MTCNN to reduce resource usage. 152 153 - **Advanced Logging**: 154 - Implement **detailed logging** of all detections (objects, faces, audio) for **post-exam analysis**. 155 156 - **Eye-Gaze Tracking**: 157 - Detect off-screen glances **more accurately**. 158 159 - **Keystroke Dynamics**: 160 - Monitor **typing patterns** to detect unusual behavior. 161 162 - **Noise Filtering**: 163 - Enhance audio detection with **advanced noise filtering** for better performance in noisy environments. 164 165 --- 166 167 ## 📜 License 168 169 This project is licensed under the **MIT License**. See the [LICENSE](LICENSE) file for details. 170 171 --- 172 173 ## 🤝 Contributing 174 175 Contributions are **welcome!** To contribute: 176 177 1. **Fork** the repository. 178 2. **Create a new branch** for your changes. 179 3. **Submit a pull request** with appropriate documentation. 180 181 Ensure your code follows the **project’s coding style**. 182 183 --- 184 185 ## 📧 Contact 186 187 For questions, suggestions, or collaboration, feel free to reach out: 188 189 - **Email**: vsm52125@gmail.com 190 - **GitHub Issues**: Open an **issue** in this repository. 191 192 --- 193 194 ## 🙏 Acknowledgments 195 196 - [YOLOv5](https://github.com/ultralytics/yolov5) → Object detection. 197 - [MTCNN](https://github.com/ipazc/mtcnn) → Face detection. 198 - [YAMNet](https://tfhub.dev/google/yamnet/1) → Audio detection. 199 - [PyQt5](https://www.riverbankcomputing.com/software/pyqt/) → UI development. 200 - **The open-source community** for invaluable resources and tools. 201 202 --- 203 204 *Built with ❤️ to ensure fairness in online education.* 205