Metricam: Fast and Reliable Social Distancing Analysis in Online Security Cameras

Distance measurements taken from 2D camera images are subject to the correct estimation of the camera’s perspective, that is, the spatial mapping from 2D points imaged by a camera to the correspondent 3D ones in the real world. Current solutions to solve this 3D reconstruction are either dependent on the estimation of vanishing points through the detection of straight lines on targeted images or by employing sophisticated sensors and deep learning algorithms, which require expensive training on huge annotated datasets. Nevertheless, none of those approaches provide the required level of precision and accuracy for social distancing evaluation. In this paper we present Metricam, a real-time lightweight software system for security cameras that computes a 2D to 3D mapping using computational geometry and uses the DBSCAN clustering algorithm to evaluate social distancing evaluation. With Metricam, we have been able to identify several places prone to agglomeration inside the Butantã campus of the University of São Paulo, and provide the local authorities with valuable information to fight off the pandemic.