MASK DETECTION
I used MobileNetV2 and OpenCV to explore face-mask classification on images, live video and small devices.
Prototype- Role
- Prototype builder
- Duration
- Earlier project
- Team
- Independent project
- Status
- Prototype
Choose the model with the device in mind
A classifier that works on a still image is only the first step toward a live camera prototype. I wanted to explore the same task on limited hardware, where the choice of model matters to whether the pipeline fits the device at all.
I used transfer learning with MobileNetV2, a compact pretrained architecture, so I could work on the classification task without training an image model from scratch. I used OpenCV to process still images and live video. I also explored what deployment could look like on Raspberry Pi and Google Coral. That brought the device into the work early, while model and pipeline choices were still open.
Decide what would count as a field test
This work stayed at prototype scope, with no measured field accuracy, public deployment or operational use. A field test would need to answer two questions: when do changes in lighting and camera angle make predictions unreliable, and can the chosen device process frames fast enough to keep up with the camera? Those questions remain open.
Working through a similar decision?
Talk to me about your project