Visual recognition for urban traffic data retrieval and analysis in major events using convolutional neural networks

Area of Science:

• Computer Vision
• Traffic Engineering
• Machine Learning

Background:

• Effective management of major events relies on accurate traffic data.
• Traditional traffic data collection is often costly and time-consuming.
• Computer vision offers a cost-efficient and timely alternative.

Purpose of the Study:

• To develop a framework for extracting traffic volume and intersection turning patterns from video data.
• To evaluate the accuracy and influencing factors of the proposed computer vision framework.
• To compare framework performance with real-world traffic data from major events.

Main Methods:

• Utilized a convolutional neural network (CNN) model for vehicle detection and tracking.
• Employed homographic projection to map vehicle data onto a real-scale map.
• Validated the framework using manually labeled video data and real-world traffic reports.

Main Results:

• Achieved robust traffic volume count accuracy up to 96.91%.
• Identified optimal camera placement parameters (pixel size > 2343, angle < 22°) for improved accuracy.
• Successfully reproduced traffic volume trends from event data and introduced a new intersection turning pattern metric.

Conclusions:

• The proposed framework provides an accurate and efficient method for traffic data analysis.
• Optimal camera setup is crucial for maximizing the performance of computer vision-based traffic monitoring.
• The framework demonstrates potential for real-world traffic management and event planning.