A Novel Vehicle Classification Using Embedded Strain Gauge Sensors.

Area of Science:

• Civil Engineering
• Transportation Engineering
• Machine Learning

Background:

• Traffic management systems require accurate vehicle classification.
• Traditional methods struggle with overlapping vehicle patterns and measurement errors.
• Machine learning offers a robust solution for complex classification tasks.

Purpose of the Study:

• To develop and validate a new vehicle classification system using pavement strain data.
• To employ machine learning, specifically Support Vector Machines (SVMs), for accurate vehicle identification.
• To enhance traffic management through reliable and automated vehicle classification.

Main Methods:

• A prototype system with five embedded strain sensors was developed to measure dynamic pavement strain.
• Vehicle parameters like wheelbase and number of axles were derived from strain data.
• Support Vector Machines (SVMs) were utilized for pattern recognition and classification of five vehicle types.
• Both one-against-all and one-against-one SVM algorithms were tested on single and multiple sensor data.

Main Results:

• The SVM-based approach successfully classified vehicles into five categories based on Federal Highway Administration guidelines.
• Classification accuracy was comparable between single and multiple sensor data using SVMs.
• Multiclass SVM fusion of multiple sensor data significantly improved classification results over single sensor data.

Conclusions:

• The developed strain-sensing system with SVMs provides a reliable method for vehicle classification.
• Machine learning, particularly SVMs, effectively overcomes limitations of traditional threshold-based classification.
• This technology has the potential to significantly improve traffic management efficiency and accuracy.