Face Mask Detection with Alarm using Machine Learning
A published computer-vision system combining CNN-based mask classification, YOLOv3 social-distance analysis, and real-time visual and audible alerts.

Published Research
Face Mask Detection with Alarm using Machine Learning was published in the International Research Journal of Modernization in Engineering, Technology and Science (IRJMETS), Volume 05, Issue 03, March 2023. I developed the project with Ganesh and Velvizhi at Sathyabama Institute of Science and Technology.
The goal was to move beyond a single image classifier and build a practical, real-time monitoring pipeline. The resulting system checks whether people are wearing masks, evaluates spacing between detected people, marks violations on the video feed, and activates an alarm when intervention is needed.
System Architecture
A camera feed is processed frame by frame with OpenCV. Faces are passed to a MobileNet-based convolutional neural network built with TensorFlow and Keras, which classifies each face as mask or no mask. In parallel, YOLOv3 detects people in the scene. Euclidean-distance calculations then estimate whether detected people are maintaining appropriate separation.
Mask classification
MobileNet-based CNN classifies detected faces in real time.
Person detection
YOLOv3 identifies people within each video frame.
Distance analysis
Euclidean distance flags unsafe proximity between detections.
Alert response
Red bounding boxes and an audible alarm identify non-compliance.
Engineering Contribution
The key contribution is the integration of two computer-vision tasks into one operational workflow. Mask classification alone cannot identify crowding, while person detection alone cannot assess mask compliance. By combining both, the project produces a more useful safety signal from a standard live camera feed. MobileNet was selected to keep inference lightweight enough for real-time use.
This work strengthened my practical understanding of dataset preparation, model integration, video-processing pipelines, object detection, and the trade-offs required when moving machine-learning models from notebooks into responsive applications.
Read the Paper
The complete paper is available through its DOI and the original IRJMETS publication archive.
From the blog
View all posts
From Security Alerts to Clearer Decisions: My Generative AI Learning Journey
What completing IBM’s cybersecurity AI course taught me about useful automation, evidence, and keeping human judgment at the center of security work.


Beyond Knowing the Answer: Building Sharper Cybersecurity Judgment
My reflections on IBM’s Security+ and CySA+ assessment course, and the value of turning technical knowledge into clear, defensible decisions.
