Advanced Computing in the Age of AI | Monday, July 22, 2024

Researchers Use Public Camera Footage to Assess Social Distancing Around the World 

Across the world, measures like social distancing and mask-wearing continue to be encouraged in efforts to stem the spread of COVID-19. Individuals’ actual, real-world behavior, however, varies wildly from region to region, and rates of COVID-19 infection have fluctuated accordingly. Now, researchers from Purdue University are using footage from public cameras around the world to assess how people in those regions are actually behaving.

These cameras, of course, already exist: live feeds of places like Times Square and Shibuya Crossing that are popular destinations for large amounts of foot traffic. The Purdue researchers built a website that pooled these resources – around 30,000 cameras spanning more than 100 countries – with the aim of using the data to automatically assess social distancing (or the lack thereof) in those areas over time. The tool is a subset of the preexisting Continuous Analysis of Many CAMeras (CAM2) tool developed by the same lab at Purdue in 2016, which itself constitutes the largest camera network in the world with more than 120,000 accessible cameras.

The tool has been pooling footage for social distancing analysis since March, archiving image data and video data about every ten minutes and sending that data to cloud datacenters for processing. Using that footage, the researchers are applying deep learning techniques to classify scenes, estimate crowd densities, detect various objects and estimate the distances between individuals (no form of facial recognition is being used, keeping the project privacy-conscious). The project is supported by a grant from the National Science Foundation (NSF) and a computing and storage allocation on the Cooley cluster at the Argonne Leadership Computing Facility (ALCF) at Argonne National Laboratory.

“Researchers already have the tools they need to analyze human behavior from video and photos, but this behavior can vary significantly depending on the context or culture of a place. We need extensive data to get those detailed insights, and this site provides that data,” said Yung-Hsiang Lu, a professor of electrical and computer engineering at Purdue.

With the tool in-hand, Lu’s lab is working to determine how policies have affected crowd sizes and foot traffic over time. 

“How have people responded to policy changes?” Lu asked. “Were there sudden increases of crowds when the restrictions lifted, or were there gradual increases? Are there obvious patterns by countries or regions? These are the types of questions we hope to answer.”

The researchers have also made the camera footage available publicly at, hoping to provide an ongoing resource for their fellow researchers and policymakers.