WorkDiCo: Workable Disease Control

By Ishan Gupta


When the number of COVID-19 cases started escalating, most nations went into a lockdown; at some point, lifting or partially easing the lockdown is inevitable due to numerous factors. Places like supermarkets, general stores, food retailers, local sellers and essential item stores (here, collectively referred to as “stores”), witnessing unavoidable crowds, could become hotspots for the spread of the disease; a recent study by four Finnish research organizations confirms this claim. WorkDiCo provides a Computer Vision and ML-enabled solution to spread customers across various locations and times to avoid crowds, encourage distancing and ensure safe visits.


WorkDiCo includes the following three components-

  1. Configurable and Installable Tools (CITs)
    It includes three tools meant to be integrated with CCTVs and local systems in stores and configured according to positioning of cameras.
    • Mask Detector
      It detects faces in real time and predicts if they are wearing a mask (with 95.5% accuracy). It includes a program which uses the open-source facial-recognition library to detect faces and a Machine Learning model, programmed in Python using Keras and TensorFlow, to predict the output for the detected faces. The model uses Convolutional Neural Networks; it has been trained on a self-compiled dataset of masked and unmasked faces.
    • People Counter
      It keeps track of the number of people present inside the store. It uses OpenCV Computer Vision library to track people entering and leaving the store in real time.
    • Physical Distancing Violation Detector (PDVD)
      It uses OpenCV and MobileNet-SSD model to detect people. It then iterates through all sets of two people and for each set, it finds the distance between the two using their three-dimensional coordinates; the depth is calculated on the basis of the ratio of the pixel height to the average human height and the focal length. If the distance is less than a minimum threshold, it is counted as a violation.
Accuracy vs. Epochs
  1. Cloud Server
    For every store, the CITs send the following metrics to a local system which constantly updates them on a server.
Mask DetectorNumber of unmasked enterers; absolute value provides a more accurate idea of risk involved.
People CounterPercentage of store filled; store capacity calculated using area and staff.
PDVDNumber of violations per minute, calculated by averaging violations for a larger window.
  1. Mobile Application
    The WorkDiCo application finds a list of accessible stores based on the user’s location, Maps API, filters like products’ categories, etc. It retrieves the aforementioned metrics from the server and processes them to rank the list of accessible stores in order of safety to suggest the most ideal (safe and uncrowded) options. It provides additional features like scheduling visits, sending shopping-lists for pre-packaging and (for managers) reporting abnormal behaviour and requesting reconfiguration.


WorkDiCo can help in optimizing essential public store visits for a post-coronavirus-lockdown world and in dealing with disease outbreaks, in general. The application utilizes metrics received from tools installed in public places to effectively make the users’ visits compliant with safe distancing practices.

External Links​

Supplemental Material (Application Demonstration Video) 

Mask Detector Code (Python, TensorFlow, Keras)  

People Counter Code (Python, OpenCV) 

Physical Distancing Violation Detector Code

(Python, OpenCV, MobileNet-SSD) 

About the author:

Ishan Gupta is a high school senior from India with a keen passion for entrepreneurship, programming and using technology for positivity. He is an academic researcher, a pianist and a passionate speaker with a flair for leadership and innovation. Being a technology enthusiast, he started a business which helps clients leverage modern technologies to boost growth. He is deeply passionate about social entrepreneurship and heads a community service organization working with under-resourced schools.

About The Author

Co-President, Harvard Tech Review

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

You don't have permission to register