A deep learning-based approach for axle counter in free-flow tolling systems

Area of Science:

• Transportation Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Toll plazas create traffic bottlenecks, hindering efficient mobility.
• Automating toll collection is crucial for improving traffic flow and operational efficiency.

Purpose of the Study:

• To develop an automated axle counter for a free-flow toll collection system.
• To enhance vehicle identification and charging processes at toll plazas.

Main Methods:

• Utilized computer vision algorithms and the You Only Look Once (YOLO) model for vehicle wheel and axle detection.
• Compared YOLOv5, YOLOv6, YOLOv7, and YOLOv8, with YOLOv5m showing the best performance.
• Developed a passage manager and image reconstruction module for complete vehicle imaging and axle counting.

Main Results:

• The YOLOv5m model achieved high precision (99.40%) and recall (98.20%) in identifying vehicle wheels.
• The system accurately counts vehicle axles and identifies vehicle passage.
• Successfully generated appropriate charges based on axle count.

Conclusions:

• The developed automated axle counter effectively contributes to a free-flow toll collection system.
• This technology enhances transportation mobility by reducing toll plaza congestion.
• Computer vision, specifically YOLO models, offers a viable solution for automated vehicle assessment in tolling applications.