Since April, computer scientists at Purdue University have been using an online database of public cameras to track compliance with social-distancing guidelines, collecting roughly 0.5 terabytes of data each week.
The team has processed more than 10.4 million webcam images through deep learning neural networks that automatically detect objects and distinguish them from people.
The algorithms draw bounding boxes around people, then calculate their distance from one another and whether they are practicing social distancing.
These observations showed that such practices are being followed to a degree, with both crowd densities and distancing lower after authorities imposed lockdowns, and higher when those restrictions were relaxed.
Purdue's Isha Ghodgaonkar thinks monitoring movements via cameras is more effective than using location-tracking data from Google and Apple, as "location tracking might be slightly biased because it's dependent on people opting in."
From New Scientist
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA