Inter-Homines: Distance-Based Risk Estimation for Human Safety

Abstract

In this document, we report our proposal for modeling the risk of possible contagiousity in a given area monitored by RGB cameras where people freely move and interact. Our system, called Inter-Homines, evaluates in real-time the contagion risk in a monitored area by analyzing video streams: it is able to locate people in 3D space, calculate interpersonal distances and predict risk levels by building dynamic maps of the monitored area. Inter-Homines works both indoor and outdoor, in public and private crowded areas. The software is applicable to already installed cameras or low-cost cameras on industrial PCs, equipped with an additional embedded edge-AI system for temporary measurements. From the AI-side, we exploit a robust pipeline for real-time people detection and localization in the ground plane by homographic transformation based on state-of-the-art computer vision algorithms; it is a combination of a people detector and a pose estimator. From the risk modeling side, we propose a parametric model for a spatio-temporal dynamic risk estimation, that, validated by epidemiologists, could be useful for safety monitoring the acceptance of social distancing prevention measures by predicting the risk level of the scene.

Monitoring social distancing with single image depth estimation

“…Since the system has to work in new environments, existing strategies [20], [16], [19] rely on a preliminary calibration phase. Specifically, [20], [16] explicitly leverage the predetermined homography between camera and ground plane to determine the 3D position on the floor of each person locating his/her feet. Additionally, [16] tries to infer the feet position even when they are occluded through a deep neural network.…”

“…Specifically, [20], [16] explicitly leverage the predetermined homography between camera and ground plane to determine the 3D position on the floor of each person locating his/her feet. Additionally, [16] tries to infer the feet position even when they are occluded through a deep neural network. Differently, [19] given an approximated knowledge of the orientation of the camera with respect to the ground plane, by assuming a fixed parameter of the human body (specifically, the constant height of 1.70 m) inferred by a deep network, detects social distancing violations by leveraging the heuristically determined torso length of each person.…”

“…interpersonal distances by means of homography computation (a) finding the position of each person as the center of the side of the bounding box touching the floor (e.g. [16], [20]) or (b) applying the method by Aghaei et al [19], automatically detecting inter-personal distances lower than 2m.…”

“…As alternative, solutions based on computer vision [16], [17], [18], [19], [20] to determine the visual social distance [21] are potentially less intrusive. They do not require explicit commitment by the people to carry mobile/wearable devices and can be adopted using standard cameras deployed for video surveillance purposes.…”

“…other devices is challenging, most approaches exploit known properties of the scene and context under examination. For instance, it is often common to leverage homography-based localization [16], [19], [18], [20] by estimating the ground plane over which people walk or by considering the overlap between detected people bounding boxes [17]. Although this allows to potentially monitor social distancing from single, uncalibrated images [19], it makes some strong assumptions on the framed scenes.…”