EUVIMEDEUVIMED entdecken
Crossref

TempoBEV: explicit motion modeling and velocity-aware fusion for roadside monocular 3D object detection

Haoyun Zhang, Wen Yang, Aiping Pang, Qianchuan Zhao

Measurement Science and Technology · 19.06.2026 · 37 (25) · 255406

DOI: 10.1088/1361-6501/ae7aa6 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 24.06.2026. Verfasst von Haoyun Zhang und 3 weiteren Autor:innen. Die Arbeit nennt 24 Literaturangaben. Gefördert von: Guizhou Provincial Department of Science and Technology, Key Programme.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In roadside monocular 3D detection, existing bird’s-eye view (BEV) methods only leverage single frames, overlooking the inter-frame geometric consistency inherent in fixed-camera continuous observations. This neglected temporal cue leads to inaccurate localization of dynamic objects and temporally inconsistent false responses from static backgrounds. To address this, we propose TempoBEV, a temporally enhanced detection framework based on explicit motion prior modeling. Its core MotionBEV module constructs a height-hypothesis-based temporal cost volume, combined with ConvGRU and illumination-invariant features, to predict pixel-level binary motion state, discrete velocity direction, and speed magnitude. These discrete priors are then embedded into continuous semantics and injected into BEV representation via motion-state-probability-modulated deformable cross-attention. On V2X-Seq-SPD, TempoBEV achieves 73.36% Moderate 3D AP for vehicles and 73.43% for cyclists, surpassing CoBEV by 6.42 and 4.63 percentage points, respectively. Moreover, DAIR-V2X-I experiments provide preliminary transferability evidence that motion priors pretrained on SPD can serve as frozen semantic priors.

Förderung

  • Guizhou Provincial Department of Science and Technology
  • Key Programme

Zitieren

APA

Zhang, H., Yang, W., Pang, A., & Zhao, Q. (2026). TempoBEV: explicit motion modeling and velocity-aware fusion for roadside monocular 3D object detection. Measurement Science and Technology, 37(25), 255406. https://doi.org/10.1088/1361-6501/ae7aa6

Vancouver

Zhang H, Yang W, Pang A, Zhao Q. TempoBEV: explicit motion modeling and velocity-aware fusion for roadside monocular 3D object detection. Measurement Science and Technology. 2026;37(25):255406. doi:10.1088/1361-6501/ae7aa6

BibTeX
@article{Zhang2026,
  author = {Zhang, Haoyun and Yang, Wen and Pang, Aiping and Zhao, Qianchuan},
  title = {TempoBEV: explicit motion modeling and velocity-aware fusion for roadside monocular 3D object detection},
  journal = {Measurement Science and Technology},
  year = {2026},
  volume = {37},
  number = {25},
  pages = {255406},
  doi = {10.1088/1361-6501/ae7aa6}
}
Kostenlos starten →