The Medical-Mask-Absence-Detection is a state-of-the-art real-time application designed for the detection of individuals not wearing medical masks in surveillance feeds. This solution is built using YOLOv8 and NVIDIA's DeepStream SDK, making it highly efficient for use in public safety and health monitoring.
Developed as an Engineering Thesis at AGH University by Kacper Bojanowski, this project integrates a seamless pipeline from webcam or RTSP inputs to a YOLOv8 model and outputs to a user-friendly web application.
(Instructions on how to install the application, including how to start the detection pipeline and interface with the web application, will be included soon)
In order to run Face Mask Detection System you must install all the needed dependencies on your Linux environment.
To execute application got to deepstream_configuration/src and run python3 deepstream_app.py
To run testing mode that will output application metrics run python3 deepstream_app.py --test
- Data processing
- Training Yolov8 custom model
- Converting it to ONNX file (Open Neural Network Exchange)
- Modifying model using GraphSurgeon for DeepStream compatibility
- Transferring it to TensorRT engine (compatibility, performance, optimization)
- Creating custom C++ parser for the Deepstream application
- Programming Deepstream Application and creating configuration file
- Analyzing results and performance
This application utilizes the following datasets for training and validation:
For support, please open an issue in the GitHub repository or contact the author directly.
NVIDIA DeepStream SDK allows for the creation and deployment of scalable AI-based video analytics applications. It provides a framework for capturing, processing, and inferencing video data, optimizing the performance on NVIDIA GPUs.
- YOLOv8: The 8th version in the YOLO series, it is a fast and accurate deep learning model for real-time object detection.
- TensorRT: NVIDIA TensorRT is a platform for high-performance deep learning inference that allows for the deployment of neural network models with optimized latency and throughput.
This project is made available under the MIT License.
- Kacper Bojanowski