Abstract
Accurate localization is a critically important issue for autonomous vehicles as it is closely related to the safety and efficiency of autonomous driving. However, current technologies for autonomous vehicle localization face many challenges. To provide accurate and robust localization services to autonomous vehicles, we propose a novel solution by employing a newly designed pavement marking. This marking operates on color contrast, temperature contrast, and binary code with some special features. We also trained and customized an object detector based on a deep learning model: YOLOv5, and integrated it with the decoding algorithm. The localization system is capable of running at a steady frame rate of more than 50 FPS. Road trials up to 80 km/h were conducted, and satisfactory results confirmed the feasibility and robustness of the localization system. Specifically, with a common onboard camera, more than four continuous frames can be detected and decoded correctly when the speed is slower than 30 km/h. At least one frame can be detected and decoded correctly at a higher speed (i.e., 30 - 50 km/h). With a high-speed camera, more than 18 frames can be detected and decoded even at 80 km/h. The findings suggest that the specially designed road marking and associated algorithms can provide a viable and economical option for accurate localization of autonomous vehicles. The performance of the system has potentials for further improvement by using better hardware such as faster CPUs, GPUs, and thermal imaging techniques.
Original language | English |
---|---|
Article number | 9780215 |
Pages (from-to) | 22290-22300 |
Number of pages | 11 |
Journal | IEEE Transactions on Intelligent Transportation Systems |
Volume | 23 |
Issue number | 11 |
DOIs | |
Publication status | Published - 1 Nov 2022 |
Keywords
- Accurate localization
- binary code
- object detection
- pavement marking
ASJC Scopus subject areas
- Automotive Engineering
- Mechanical Engineering
- Computer Science Applications