Classification and Segmentation of Longitudinal Road Marking using Convolutional Neural Networks for Dynamic
Chanjun Chun1, Taehee Lee1, Sungil Kwon2
1Future Infrastructure Research Center, Korea Institute of Civil Engineering and Building Technology (KICT), Goyang 10223, Korea.
Sensors (Basel, Switzerland)
|October 1, 2020
Summary
This study introduces a dynamic method using luminance cameras and convolutional neural networks (CNNs) to estimate road marking retroreflection. The approach enhances road safety and sustainability by accurately measuring reflection properties in real-time.
Area of Science:
- Road safety engineering
- Computer vision
- Materials science
Background:
- Road markings are critical infrastructure requiring specific standards for luminous contrast, known as retroreflection.
- Effective management of retroreflection is vital for improving road safety and sustainability.
Discussion:
- A novel dynamic retroreflection estimation method for longitudinal road markings is proposed.
- This method utilizes a luminance camera and advanced convolutional neural networks (CNNs).
Key Insights:
- A classification and regression CNN model validates the accurate acquisition of road marking images.
- A segmentation model precisely delineates road markings and reference plates within captured images.
Outlook:
- The method was validated through dynamic, real-world driving measurements, demonstrating its effectiveness.
- This research contributes to more sustainable and safer road infrastructure management.
