Real-time detection of road surface friction coefficient: A new framework integrating diffusion model and Transformer in Transformer algorithms
The real-time road surface friction coefficient (RSFC) is a critical parameter for evaluating skid resistance and making safe driving decisions in driver assistance systems and autonomous vehicles, especially under adverse weather conditions. RSFC estimation depends on the interaction between the ro...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-02-01
|
| Series: | Alexandria Engineering Journal |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1110016824014170 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The real-time road surface friction coefficient (RSFC) is a critical parameter for evaluating skid resistance and making safe driving decisions in driver assistance systems and autonomous vehicles, especially under adverse weather conditions. RSFC estimation depends on the interaction between the road surface and tires. However, accurate estimation is challenging due to varying road environments and sensor errors that can cause significant distortions. To obtain high-accuracy RSFC, this study proposes a novel real-time RSFC detection method that integrates a diffusion model with the Transformer-in-Transformer(TNT) model to detect RSFC from vehicle video pictures. The method consists of three steps. First, we created labeled friction coefficient image datasets representing asphalt concrete surfaces under four moisture conditions. Second, we used a diffusion model to enhance the dataset, increasing sample diversity. Finally, we trained a TNT model on the extended dataset to recognize friction coefficients. The approach was tested across various datasets and compared to four state-of-the-art (SOTA) methods. The results show that the proposed method significantly improves accuracy, achieving a 22.89% increase compared to the unenhanced dataset and a 5.59% improvement over SOTA methods. The primary contribution of this study is the integration of generative artificial intelligence and computer vision algorithms to enhance RSFC recognition accuracy. Furthermore, the recognition method meets the real-time performance requirements, processing frames in just two milliseconds. This method can be an effective tool for perceiving road surface environmental parameters and holds significant value in improving driving safety under adverse weather conditions. |
|---|---|
| ISSN: | 1110-0168 |