A Vision-based Social Distancing and Critical Density Detection System for COVID-19 (2007.03578v2)

Abstract: Social distancing has been proven as an effective measure against the spread of the infectious COronaVIrus Disease 2019 (COVID-19). However, individuals are not used to tracking the required 6-feet (2-meters) distance between themselves and their surroundings. An active surveillance system capable of detecting distances between individuals and warning them can slow down the spread of the deadly disease. Furthermore, measuring social density in a region of interest (ROI) and modulating inflow can decrease social distancing violation occurrence chance. On the other hand, recording data and labeling individuals who do not follow the measures will breach individuals' rights in free-societies. Here we propose an AI based real-time social distancing detection and warning system considering four important ethical factors: (1) the system should never record/cache data, (2) the warnings should not target the individuals, (3) no human supervisor should be in the detection/warning loop, and (4) the code should be open-source and accessible to the public. Against this backdrop, we propose using a monocular camera and deep learning-based real-time object detectors to measure social distancing. If a violation is detected, a non-intrusive audio-visual warning signal is emitted without targeting the individual who breached the social distancing measure. Also, if the social density is over a critical value, the system sends a control signal to modulate inflow into the ROI. We tested the proposed method across real-world datasets to measure its generality and performance. The proposed method is ready for deployment, and our code is open-sourced.

• The paper presents a vision-based AI system using deep CNNs like Faster R-CNN and YOLOv4 to detect human presence and monitor distancing.
• It computes inter-personal distances and defines a 'critical density' threshold to trigger audio-visual alerts when conditions are violated.
• Empirical evaluations in urban, mall, and train station settings demonstrate robust, privacy-aware performance in diverse environments.

A Vision-Based Social Distancing and Critical Density Detection System for COVID-19

In response to the COVID-19 pandemic, the paper introduces a vision-based artificial intelligence system designed to monitor social distancing and manage crowd density in real time using a monocular camera. Developed by researchers from The Ohio State University, the system is intended to identify social distancing violations and provide non-intrusive alerts to increase compliance without compromising personal privacy. This technology presents a novel approach to public health monitoring during a crisis by relying on intelligent object detection methods integrated with ethical design considerations.

The system utilizes deep convolutional neural networks (CNNs) for object detection, specifically targeting human presence within a region of interest (ROI). Pedestrian detection is accomplished with models like Faster R-CNN and YOLOv4, which transform image data into actionable spatial coordinates mapped onto a bird's-eye view in the physical world. The system calculates inter-personal distances and alerts individuals through audio-visual cues when social distancing guidelines are breached.

A significant contribution of this paper is the definition and utilization of a "critical social density" threshold—the maximum crowd density that ensures low probabilities of social distancing violation. By defining overcrowding in statistical terms, the system can autonomously manage inflow into monitored areas, advising when the density is at levels likely to lead to breaches.

The empirical evaluation of the system is conducted through three case studies, notably using datasets from an urban center, an indoor mall, and a major train station. These evaluations reveal consistent detection accuracy across varied environments despite challenges such as occlusion in dense areas. The implementation demonstrates that existing object detection networks achieve real-time performance suitable for deployment in social environments without storing any privacy-sensitive data.

The results illustrate the negative correlation between crowd density and adherence to distancing rules, showcasing the effectiveness of linear regression in predicting and managing crowd density thresholds. By calculating metrics like minimum and average distances between individuals and applying regression analysis to density data, the system effectively offers a quantitative approach to crowd management beyond simple surveillance.

Overall, this paper contributes a well-defined technological blueprint for public health management systems emphasizing privacy, ethics, and open-source accessibility. Future research directions could include examining group behaviors in pedestrian dynamics or extending this methodology to multi-camera systems for more comprehensive spatial coverage. The implications of the research extend to theoretical domains in computational ethics and practical implementations in AI-driven public safety and compliance monitoring systems.