Real-Time and Accurate UAV Pedestrian Detection for Social Distancing Monitoring in COVID-19 Pandemic

Area of Science:

• Computer Vision
• Artificial Intelligence
• Public Health

Background:

• Coronavirus Disease 2019 (COVID-19) necessitates effective public health interventions.
• Social distancing is a critical non-pharmaceutical measure to mitigate virus transmission.
• Unmanned Aerial Vehicles (UAVs) offer a flexible platform for remote monitoring applications.

Purpose of the Study:

• To develop a lightweight pedestrian detection network for real-time social distancing monitoring using UAV imagery.
• To accurately detect pedestrians via human head detection for distance calculation.
• To evaluate the network's performance against existing models and in real-world conditions.

Main Methods:

• Proposed a lightweight pedestrian detection network based on PeleeNet backbone.
• Incorporated multi-scale features and spatial attention to enhance small object detection (human heads).
• Conducted precision calibration for image-to-real-world coordinate transformation.

Main Results:

• Achieved 92.22% Average Precision (AP) and 76 Frames Per Second (FPS) on the Merge-Head dataset, outperforming YOLOv3 and SSD.
• Demonstrated high precision pedestrian detection on UAV images (88.5% AP, 75 FPS) using the UAV-Head dataset.
• Ablation studies confirmed the significant contribution of multi-scale features and spatial attention.

Conclusions:

• The proposed network enables accurate, real-time pedestrian detection and social distancing monitoring from UAV images.
• The method provides a reliable solution for public health surveillance and crowd management.
• The integration of advanced deep learning techniques enhances the effectiveness of UAV-based monitoring systems.