The project is about the development of an algorithm that could help in controlling the use of masks in public places by the analysis of video footage.
The code constitutes an implementation of the approach proposed by Joshi et al.
- OpenCV
- Numpy
- Tensorflow
Detect person faces from images taken in a wild environment with MTCNN, a CNN that exploits multi-task learning to integrate face detection and face alignment.
In FaceMaskDetection_main.ipynb, MTCNN was evaluated on Face Mask Dataset that contains 924 images of people with and without mask.
| Precision | Recall | F1 |
|---|---|---|
| 82.8% | 74.4% | 78.4% |
Predict the presence of masks with MobileNetV2, a network architecture that performs feature extraction for object detection and segmentation.
In MaskDetection_training.ipynb the cited network was trained on Face Mask Detection ~12K Images Dataset that contains only faces of people with and without masks.
In FaceMaskDetection_main.ipynb the trained MobileNetV2 model was used evaluated on the Face Mask Dataset.
| Precision | Recall | F1 | |
|---|---|---|---|
| WithMask | 75% | 93% | 83% |
| WithoutMask | 91% | 69% | 78% |