alessandro-marcantoni / mask-detection-social-distancing-var

Mask Detection & Social Distancing

Visione Artificiale e Riconoscimento

This repository contains the project developed for the above course. The two main parts are:

• Mask Detection: this solution detects people's faces in pictures, ideally taken in a small room, and determines whether they are wearing a face mask correctly, not correctly or not at all.
• Social Distancing: this solution detects people in pictures, ideally taken in a small room, and checks whether they are abiding by the social distancing rules or not.

Requirements

In order to use the solutions above, you will need to have python >= 3.7.12 and pip >= 22.0.3 installed on your machine. The required libraries can be easily downloaded and installed by running the following command:

pip install -r requirements.txt

Mask Detection

Inside the mask-detection directory you will find the following sources:

• training.ipynb: this is the notebook used to perform the training of the classifiers for the mask classification task.

• performance_eval.ipynb: this is the notebook used to test the performances of the entire solution.

• facemask_detection.py: this is the script that performs the inference using both the retinaface model and the trained classifier; it takes two command line parameters:

  • The path of the video to be inferred or "webcam" to infer the webcam video stream in real time.
  • Either "ssd" or "svm" depending on which classifier you want to use.

  python facemask_detection.py <path/to/video> <model>
  

Resources

You will need to download and unzip the facemask-detection-resources.zip archive in the mask-detection folder. It contains:

• The retinaface model for face detection;
• The mobilenet model for mask classification;
• The SVC model for mask classification;
• The dataset used for training;
• The video and labels used for test.

Social Distancing

Inside the social-distancing directory you will find the following sources:

• social_distancing_eval.ipynb: this is the notebook used to test the performances of different neural networks and choose the best one.

• social_distancing_matrix.py: this is the script used to generate the transformation perspective matrix which is fundamental for the social distancing verification task.

• social_distancing_verification.py: this is the script that performs the inference using the best model; it takes one command line parameter:

  • The path of the video to be inferred.

  python social_distancing_verification.py <path/to/video>
  

Resources

You will need to download and unzip the social-distancing-resources.zip archive in the social-distancing folder. It contains:

• The centernet-resnet50v1-fpn used for inference;
• Video example of people in a room to try the inference script;
• Images and labels used for test;
• Transformation matrix for both the video example and test set.