DeepSOCIAL: Social Distancing Monitoring and Infection Risk Assessment in COVID-19 Pandemic (2008.11672v3)

Abstract: Social distancing is a recommended solution by the World Health Organisation (WHO) to minimise the spread of COVID-19 in public places. The majority of governments and national health authorities have set the 2-meter physical distancing as a mandatory safety measure in shopping centres, schools and other covered areas. In this research, we develop a hybrid Computer Vision and YOLOv4-based Deep Neural Network model for automated people detection in the crowd in indoor and outdoor environments using common CCTV security cameras. The proposed DNN model in combination with an adapted inverse perspective mapping (IPM) technique and SORT tracking algorithm leads to a robust people detection and social distancing monitoring. The model has been trained against two most comprehensive datasets by the time of the research the Microsoft Common Objects in Context (MS COCO) and Google Open Image datasets. The system has been evaluated against the Oxford Town Centre dataset with superior performance compared to three state-of-the-art methods. The evaluation has been conducted in challenging conditions, including occlusion, partial visibility, and under lighting variations with the mean average precision of 99.8% and the real-time speed of 24.1 fps. We also provide an online infection risk assessment scheme by statistical analysis of the Spatio-temporal data from people's moving trajectories and the rate of social distancing violations. The developed model is a generic and accurate people detection and tracking solution that can be applied in many other fields such as autonomous vehicles, human action recognition, anomaly detection, sports, crowd analysis, or any other research areas where the human detection is in the centre of attention.

• The paper introduces an AI-driven system employing YOLOv4 for high-accuracy human detection, achieving 99.8% mAP at 24.1 fps.
• It uses inverse perspective mapping and SORT tracking to accurately estimate interpersonal distances in varied indoor and outdoor settings.
• The study demonstrates practical COVID-19 risk assessment by identifying high violation zones, thereby informing targeted public health interventions.

An Analytical Overview of "DeepSOCIAL: Social Distancing Monitoring and Infection Risk Assessment in COVID-19 Pandemic"

In the wake of the COVID-19 pandemic, public health authorities globally have emphasized social distancing as a critical measure to curb the virus's spread. Traditional monitoring of these measures has proven labor-intensive and prone to error, particularly in densely populated settings. In response, the paper titled "DeepSOCIAL: Social Distancing Monitoring and Infection Risk Assessment in COVID-19 Pandemic" by Rezaei and Azarmi introduces an AI-driven approach for real-time monitoring of social distancing utilizing a hybrid computer vision model.

The proposed solution, DeepSOCIAL, integrates the YOLOv4 deep neural network model with computer vision techniques, addressing the necessity for an efficient and automated human detection and tracking system in indoor and outdoor environments. The emphasis is on integrating the system with existing CCTV infrastructures, thus promoting broad applicability without the necessity for specialized hardware.

Methodological Framework

1. Model Architecture and Datasets: The paper leverages the strengths of YOLOv4 for object detection, which is optimized here for single-class (human) detection. The model was trained using two comprehensive datasets—MS COCO and Google Open Images—totaling over 3.7 million annotated instances, to ensure robustness against varied scenarios of human presentation, including occlusions and varying lighting conditions.
2. System Performance Evaluation: The DeepSOCIAL system underwent rigorous performance evaluations on the Oxford Town Centre dataset, encompassing approximately 150,000 instances of human detection. The model demonstrated a mean average precision of 99.8% and operated at a real-time speed of 24.1 fps. This performance surpasses several state-of-the-art methodologies, affirming the model's capability in complex environments.
3. Distance Estimation and Tracking: To mitigate the challenges of depth perception in single-camera systems, the researchers employ inverse perspective mapping (IPM), which corrects for perspective distortion in 2D images, transforming these into a bird's-eye view (BEV) for accurate interpersonal distance measurements. Additionally, the SORT algorithm facilitates real-time tracking of individuals, ensuring consistent ID assignment even amidst temporary occlusions.
4. Infection Risk Assessment: Beyond detection and tracking, DeepSOCIAL incorporates a statistical framework for risk assessment. By analyzing spatio-temporal data on tracked individuals, the model identifies zones with high violation rates of social distancing guidelines. These insights can be vital for policymakers aiming to redesign public spaces or enforce stricter measures in identified 'high-risk' zones.

Implications and Future Prospects

The implications of implementing DeepSOCIAL are significant, extending beyond immediate applications in combating COVID-19. By enhancing the precision of pedestrian detection and social distancing monitoring, such technology can bolster various domains, including autonomous vehicles, crowd management, and public safety systems. This comprehensive approach to risk assessment also suggests a prototype for future public health surveillance systems, capable of real-time data analysis and decision support.

The theoretical contributions of this paper are poised to influence ongoing developments in AI-based surveillance technologies. It offers a template for deploying deep learning within public health frameworks, highlighting a pathway towards data-driven decision-making processes in crisis scenarios.

Conclusion

DeepSOCIAL pioneers a substantial leap in automated social distancing monitoring through an integrated use of advanced neural networks and computer vision techniques. It balances high precision with real-time capabilities, fulfilling a critical need in pandemic response strategies. However, future iterations could explore additional modalities, such as multi-camera integration or thermal sensing, to enhance robustness and applicability across diverse real-world environments. The continued evolution of such tools will undoubtedly play a pivotal role in managing public health challenges, both current and forthcoming.