The Objective Assessment of COVID-19 Transmission Behaviors


Background: The H1N2 Coronovirus, commonly referred to as COVID-19, has devastated economies, education, personal well-being, and many other aspects of human life. Screening, therapeutics, and vaccines are necessary to combat the virus as are preventative measures such as mask wearing and social distancing. While most health agencies agree these basic behaviors reduce the transmission of COVID-19, there are few systematic studies on their prevalence beyond survey (self-report) research. Therefore, the current investigation utilized the observation method and video technology in a novel way to obtain information on COVID-19 transmission behaviors and potential moderators (e.g., sex, age). Methods: A wearable video device (eye glasses) was worn by a data collector and videos were continuously recorded while traversing a 2-mi long, pre-determined route located on and around a large, University campus. This was performed between 11:15 a.m. and 12:15 p.m., once per week over the course of six consecutive weeks. Data from the videos was independently extracted by two trained investigators who then reached consensus on the desired information. Results: At total of 872 people were observed in 7 h 30 min of video (~2 people/min). Most people (96%) were walking or standing/sitting, 50.9% were mask non-compliant, 45.4% were not social distancing, 27.6% were simultaneously mask and social distancing non-compliant, 8.7% touched their face, and 18.3% touched a surface. Transmission behaviors varied by demographics with white, overweight/obese males least likely to be mask-compliant (59.6%) and white, overweight/obese females least likely to practice social distancing (57.1%). Certain environments (e.g. crosswalks) were identified as “hot spots” or areas where higher rates of adverse transmission behaviors occurred. Conclusion: This study introduces a methodology for obtaining objective data on COVID-19 transmission behaviors that could easily be adapted for use by others (e.g. researchers, practitioners). Based on this localized evaluation, compliance is less than recommended for significantly impeding COVID-19 transmission. Certain human characteristics and environments were identified that could potentially be targeted by interventions promoting mask and social distancing compliance.


Drone Footage for Determining Human Movement Patterns

Wearable Video Device

Social Distancing and Mask Wearing Behaviors (data extracted using computer vision – the “boxes”)


What is LabelVision?

  • Open source data annotation software based in Python
  • Easy to install and use, clear documentation
  • Will be used to detect the movement of people and social distancing practices. Mask? No mask?
  • Creates YOLO format text files that record the position of people and mask usage.

Labeling Rules

  • Bounding boxes should be exhaustively labeled.
  • Mask boxes need to within people boxes.
  • Bounding boxes should be as tight as possible and include all visible parts. Occluded parts do not matter.

Automated Data Quality Assessment

  • Stochastic Gradient Descent algorithm to identify which mask label corresponds to which person label. Integrate a python script to ensure that:
    1. Every person has a mask label
    2. Person label encloses a mask label
    3. Pair each person label with a mask label



Identifying Masks & Linking to People

Identifying social distancing

Bounding Boxes

Stochastic Gradient Descent for Associating Masks with People