EUVIMEDEUVIMED entdecken
Crossref

Driving Behavior Classification Method Based on Fourier Transform Multimodal Fusion

Xing Sheng, Jianrong Cao, Junzhe Zhang, Zhen Wang, Zongtao Duan

IEEE Transactions on Intelligent Transportation Systems · 2025-10 · 26 (10) · 15000-15012

DOI: 10.1109/tits.2025.3579142 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Intelligent Transportation Systems, 2025-10. Verfasst von Xing Sheng und 4 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 55 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Shaanxi Special Support Program for Leading Talents in Science and Technology Innovation, Young Science and Technology Talent fund of Shaanxi, China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Driving behavior classification is an important component of advanced driver assistance system (ADAS) and plays a pivotal role in enhancing driving safety and economy. Existing driving behavior classification methods primarily rely on time-domain features or multimodal fusion techniques. However, these methods often involve many parameters and complex training processes, making them difficult to be directly applied in real-world scenarios. This study proposes a lightweight driving behavior classification method that integrates both time-domain and frequency-domain feature. The proposed method comprises two parallel branches. The first branch extracts time-domain features using a combination of Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory Network (BiLSTM). Instead of the conventional fully connected layers, the Kolmogorov-Arnold Network (KAN) is employed to reduce modeling complexity and enhance performance. The second branch captures frequency-domain features through the Discrete Fourier Transform (DFT) method. An adaptive filtering block removes high-frequency noise, and the frequency-domain features are then fused adaptively using two learnable filters—global and denoising. The time-frequency domain multimodal feature fusion module performs the final weighted fusion of these features. Extensive experiments are conducted to evaluate the proposed method based on the UAH-DriveSet. Experimental results show that our method achieves an F1 score of 98.74%, surpassing the current state-of-the-art. Moreover, to assess the robustness of our method, experiments are conducted on the Ford Stay Alert Challenge dataset, obtaining an F1 score of 97.83%. Furthermore, our method is highly efficient, with only 3.97M parameters and 0.38G FLOPS, significantly lower than existing methods. The inference speed reaches an impressive 1,257 FPS, meeting the real-time requirements in resource-constrained environments.

Förderung

  • National Natural Science Foundation of China
  • Shaanxi Special Support Program for Leading Talents in Science and Technology Innovation
  • Young Science and Technology Talent fund of Shaanxi, China
  • Fundamental Research Funds for the Central Universities, Chang'an University
  • Taishan Industrial Experts Program
  • Young Elite Scientists Sponsorship Program
  • China Association for Science and Technology

Zitieren

APA

Sheng, X., Cao, J., Zhang, J., Wang, Z., & Duan, Z. (2025). Driving Behavior Classification Method Based on Fourier Transform Multimodal Fusion. IEEE Transactions on Intelligent Transportation Systems, 26(10), 15000-15012. https://doi.org/10.1109/tits.2025.3579142

Vancouver

Sheng X, Cao J, Zhang J, Wang Z, Duan Z. Driving Behavior Classification Method Based on Fourier Transform Multimodal Fusion. IEEE Transactions on Intelligent Transportation Systems. 2025;26(10):15000-15012. doi:10.1109/tits.2025.3579142

BibTeX
@article{Sheng2025,
  author = {Sheng, Xing and Cao, Jianrong and Zhang, Junzhe and Wang, Zhen and Duan, Zongtao},
  title = {Driving Behavior Classification Method Based on Fourier Transform Multimodal Fusion},
  journal = {IEEE Transactions on Intelligent Transportation Systems},
  year = {2025},
  volume = {26},
  number = {10},
  pages = {15000--15012},
  doi = {10.1109/tits.2025.3579142}
}
Kostenlos starten →