Robust Vehicle Speed Measurement Based on Feature Information Fusion for Vehicle Multi-Characteristic Detection

Area of Science:

• Computer Vision
• Machine Learning
• Intelligent Transportation Systems

Background:

• Accurate vehicle speed measurement is crucial for traffic safety and management.
• Existing systems often struggle with robustness and accuracy under diverse conditions.
• Multi-characteristic detection offers a promising approach to enhance measurement reliability.

Purpose of the Study:

• To develop a robust vehicle speed measurement system using feature information fusion.
• To improve upon existing object detection algorithms for enhanced vehicle multi-characteristic detection.
• To design a system that meets stringent national accuracy standards.

Main Methods:

• Construction of a vehicle multi-characteristic dataset.
• Training and evaluation of seven CNN-based object detection algorithms.
• Selection and enhancement of the FPN-based YOLOv4 algorithm with an ECA channel attention module.
• Development of a speed measurement system integrating license plate, logo, and light features.

Main Results:

• The enhanced ECA-YOLOv4 algorithm demonstrated improved performance in vehicle multi-characteristic detection.
• The proposed system significantly reduced model parameter size and FLOPs (Floating Point Operations Per Second).
• Experimental results confirmed the system's speed measurement error rate is below the 6% national standard (GB/T 21555-2007).

Conclusions:

• The proposed ECA-YOLOv4 algorithm effectively fuses multi-scale features and cross-channel information for superior detection.
• The developed vehicle speed measurement system enhances accuracy and robustness.
• The system successfully meets national standards, offering a reliable solution for intelligent transportation.