COVID-Precaution-Monitor

This repository contains all the project files developed by me, during HackNagpur 2020.

Purpose:

The purpose for this project is to enable a user to monitor the index of COVID precautionary measures being followed or not. Our model is a combination of Social-Distancing-Monitor and Fask-Mask-Monitor. The model helps to monitor people violating saftey norms over video footage from CCTV cameras.

I've used YOLOv3 along with DBSCAN clustering for recognizing potential violations. A pre-trained Face Mask Classifier model (ResNet50) is used for detecting if the people are wearing face masks or not. Links to both (1)YoloV3.weights & (2)PreTrained ResNet50 Model can be found here.

Requirements:

Making a virtual enviroment is strongly suggested. Click here to know more about virtual enviroments using Anaconda.

Following python packages would be required for running the project:

• numpy
• matplotlib
• sklearn
• Pillow
• opencv-python(OpenCV)
• keras
• face-detection
• face-recognition
• tqdm

Usage:

Run the following command in your terminal:

python Covid_Precaution_Monitor.py

After a bunch of TensorFlow warnings and stats, you will be able to see a progress bar processing the input video frame-by-frame.

It would take some time to process the video(depending upon your system's specifications).

After completion of the progress bar, you would be able to see the result.mp4 file as well as the various output images in the results folder.(I have commented those lines, but you can uncomment those in the code if you wish to store these files locally.)

NOTE: You have to move/remove the result video and files(if applicable) before processing another video, otherwise it will throw an error saying "directory/file already exists".

Example Output:

Potential Improvements:

We can improve the processing speed by not saving each frame, detected persons and detected faces after each iteration locally.

• Implemented! (Merged own pull request from improvements branch)

Contact Me:

Feel free to reachout on my LinkedIn if you got any queries or need any help: