EUVIMEDEUVIMED entdecken
Crossref

Traffic Road Visibility Retrieval in the Internet of Video Things Through Physical Feature-Based Learning Network

Yuexia Wang, Linyi Zhou, Zhihuo Xu

IEEE Transactions on Intelligent Transportation Systems · 2025-03 · 26 (3) · 3629-3642

DOI: 10.1109/tits.2024.3519137 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Intelligent Transportation Systems, 2025-03. Verfasst von Yuexia Wang und 2 weiteren Autor:innen. Die Arbeit wurde bisher 6-mal zitiert (Crossref) und nennt 30 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Jiangsu Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study addresses the challenge of retrieving traffic road visibility in the Internet of Video Things (IoVT). The complexity arises from the highly ill-posed inverse problem of estimating road visibility from video frames due to indirect inference, leading to ambiguity and multiple possible solutions. Initially, this study formulates the fog effects on images and discovers that the first and second eigenvalues of the observed image matrix closely approximate those of the airlight component under fog conditions. Based on this important discovery, a four-step framework is proposed for estimating road visibility. The framework includes defining persistent scatterers, formulating a singular value decomposition (SVD)-based method for background and airlight separation, extracting crucial physical features, and designing a hybrid convolutional long short-term memory (LSTM) network for accurate visibility estimation. Specifically, the foundational step defines persistent scatterers in visual scenes, followed by a sophisticated SVD-based method for background and airlight separation. Subsequently, physical features, including eigenvalue-based entropy and persistent scatterer intensity, are computed. The framework concludes with a novel hybrid convolutional LSTM network tailored for traffic road visibility estimation. To validate the methodology, three comparative research methods are introduced: one based on the Koschmieder law, another utilizing a convolutional neural network (CNN), and a third employing a deep LSTM approach. Results indicate correlation coefficients of 0.2417, 0.3325, 0.7930, and 0.9484 for the Koschmieder law-based, CNN, deep LSTM, and the proposed method, respectively, when compared to true visibility measurements. Furthermore, the averaged root mean square estimation errors are 6539 m, 9095 m, 1600 m, and 681 m for the Koschmieder law-based, CNN, deep LSTM, and the proposed method, respectively. The data and code for this study are available at https://github.com/Z-H-XU/Benchmark-Visibility to facilitate reproducibility within the research community.

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Jiangsu Province

Zitieren

APA

Wang, Y., Zhou, L., & Xu, Z. (2025). Traffic Road Visibility Retrieval in the Internet of Video Things Through Physical Feature-Based Learning Network. IEEE Transactions on Intelligent Transportation Systems, 26(3), 3629-3642. https://doi.org/10.1109/tits.2024.3519137

Vancouver

Wang Y, Zhou L, Xu Z. Traffic Road Visibility Retrieval in the Internet of Video Things Through Physical Feature-Based Learning Network. IEEE Transactions on Intelligent Transportation Systems. 2025;26(3):3629-3642. doi:10.1109/tits.2024.3519137

BibTeX
@article{Wang2025,
  author = {Wang, Yuexia and Zhou, Linyi and Xu, Zhihuo},
  title = {Traffic Road Visibility Retrieval in the Internet of Video Things Through Physical Feature-Based Learning Network},
  journal = {IEEE Transactions on Intelligent Transportation Systems},
  year = {2025},
  volume = {26},
  number = {3},
  pages = {3629--3642},
  doi = {10.1109/tits.2024.3519137}
}
Kostenlos starten →