Deep Learning-Based Congestion Detection at Urban Intersections

Area of Science:

• Computer Vision
• Artificial Intelligence
• Transportation Engineering

Background:

• Urban traffic congestion is a significant issue impacting efficiency and safety.
• Accurate real-time traffic state detection is crucial for intelligent transportation systems.
• Existing methods may lack accuracy or robustness in complex urban intersection environments.

Purpose of the Study:

• To propose a novel deep learning-based method for discriminating traffic states at urban intersections.
• To accurately detect vehicle speeds and classify traffic conditions (e.g., congestion).
• To develop a system with strong anti-interference capabilities for practical deployment.

Main Methods:

• Utilizing the You Only Look Once (YOLO) v3 algorithm for vehicle detection and localization.
• Employing the Lucas-Kanade (LK) optical flow method to calculate vehicle speeds based on YOLOv3 outputs.
• Integrating vehicle speed data with a discrimination algorithm to determine intersection traffic states.

Main Results:

• The proposed method accurately detects vehicle speeds within the selected region of interest (urban intersections).
• The traffic state discrimination algorithm effectively classifies intersection traffic conditions.
• Experimental results validate the algorithm's accuracy and strong anti-interference ability.

Conclusions:

• The deep learning-based approach provides an accurate and reliable method for traffic state discrimination at urban intersections.
• The system meets practical application requirements for intelligent transportation systems.
• This method offers a promising solution for mitigating urban traffic congestion through enhanced monitoring.