Advanced Pedestrian State Sensing Method for Automated Patrol Vehicle Based on Multi-Sensor Fusion

Area of Science:

• Robotics and Automation
• Computer Vision
• Public Health Technology

Background:

• The ongoing COVID-19 pandemic necessitates continuous monitoring of public spaces to mitigate virus transmission.
• Pedestrians in public areas remain susceptible to infection, requiring effective surveillance methods.
• Existing surveillance systems may lack the accuracy and comprehensive sensing capabilities needed for dynamic crowd monitoring.

Purpose of the Study:

• To develop and validate an advanced pedestrian state sensing method for automated patrol vehicles.
• To enhance the detection of pedestrians, estimate crowd density, and identify individuals with abnormal body temperatures.
• To reduce the risk of cross-infection in public areas through automated monitoring.

Main Methods:

• A multi-sensor fusion approach combining Euclidean clustering and YOLO V4 for pedestrian detection.
• Decision-level fusion to improve pedestrian detection accuracy.
• Multi-layer fusion for crowd density distribution calculation and estimation.
• Thermal infrared cameras for detecting body temperature in aggregated crowds.

Main Results:

• Mean accuracy of pedestrian detection increased by 17.1% compared to single-sensor methods.
• Mean error in crowd density estimation was 3.74%.
• Maximum error in body temperature detection was less than 0.8°C, enabling identification of abnormal temperature targets.

Conclusions:

• The proposed multi-sensor fusion method significantly improves pedestrian detection and crowd density estimation accuracy.
• The system effectively identifies individuals with elevated body temperatures, crucial for epidemic control.
• This automated patrol vehicle-based sensing technique offers an efficient solution for public health surveillance and epidemic prevention.