Vehicular Traffic Congestion Classification by Visual Features and Deep Learning Approaches: A Comparison

Area of Science:

• Computer Vision
• Artificial Intelligence
• Transportation Engineering

Background:

• Surveillance cameras on roads enable real-time vehicle identification for traffic flow estimation.
• Accurate traffic flow classification is crucial for detecting congestion and accidents.

Purpose of the Study:

• To conduct a comparative analysis of state-of-the-art object detectors, visual features, and classification models for traffic state estimation.
• To evaluate classic machine learning approaches against deep learning techniques for this task.

Main Methods:

• Comparison of three object detectors for vehicle identification.
• Employment of four machine learning techniques with five visual features.
• Evaluation of classic machine learning against deep learning models.

Main Results:

• Encouraging results were observed for both classic machine learning and deep learning methods.
• Deep learning models demonstrated superior performance in traffic state classification.
• Achieved 99.9% accuracy for binary classification and 98.6% for multiclass classification.

Conclusions:

• Deep learning offers the most accurate approach for automatic traffic flow classification.
• The findings support the effective implementation of deep learning for intelligent transportation systems.