An Optimized Deep Learning Approach for the Prediction of Social Distance Among Individuals in Public Places During...

Area of Science:

• Computer Science
• Artificial Intelligence
• Public Health

Background:

• Social distancing is crucial for pandemic control, yet public compliance remains a challenge.
• Existing methods lack robust monitoring systems for social distancing in public areas.
• Violations of social distancing norms have been widely observed despite government notifications.

Purpose of the Study:

• To propose an optimized deep learning model for predicting social distancing adherence in public spaces.
• To develop a system for monitoring and evaluating violations of social distancing norms.
• To enhance pandemic prevention strategies through automated compliance monitoring.

Main Methods:

• Implemented a customized deep learning model using detectron2 and intersection over union (IOU).
• Utilized Faster R-CNN for social distancing norm prediction and detectron2 for 'person class' identification.
• Conducted extensive trials comparing various state-of-the-art object detection models.

Main Results:

• The proposed R-CNN with detectron2 and Fast R-CNN model demonstrated superior performance.
• Achieved approximately 97.9% precision and an 87% recall score.
• The model effectively estimated violation-threat criteria with a calculated threshold of 0.75.

Conclusions:

• The developed deep learning model offers an efficient solution for monitoring social distancing compliance.
• This technology can significantly aid authorities in enforcing public health guidelines during pandemics.
• The model's high precision and recall indicate its potential for real-world application in public safety.