EUVIMEDEUVIMED entdecken
Crossref

Action-Semantics-Guided Z -Axis Point Cloud Reconstruction for Millimeter-Wave Radar

Wei Yin, Ling-Feng Shi, Yifan Shi

IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 8512111-8512111

DOI: 10.1109/tim.2026.3716503 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Wei Yin und 2 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In industrial indoor settings, mmWave radar often yields unstable $Z$ -axis (height) estimates because the pitch-direction virtual aperture is limited. This produces semi-ring/band-likeZdistortions and flattened point clouds, obscuring human structure and blurring limb contours, which degrades readability and downstream interpretation. The root cause is that sparse pitch virtual elements compress pitch information into tiny phase differences, making height reconstruction sensitive to noise, multipath, and calibration errors; small pitch jitter is then amplified into pronounced $Z$ -axis curvature in 3-D imaging. We propose Action-guided $Z$ -axis Reconstruction Network (ActZ-Net), which incorporates action-category priors for $Z$ -axis geometric reconstruction to restore clearer and more interpretable human 3-D structures from sparse mmWave radar point clouds with severe $Z$ -axis distortion. ActZ-Net comprises three components: 1) action-semantically adaptive normalization (Ada-ActNorm) enables differentiated activation of geometric channels across different actions, 2) $Z$ -axis anisotropic attention explicitly incorporating thezcoordinate to enhance height modelling, and 3) offset regression based on a reference point set to avoid geometric drift caused by direct coordinate generation. Experiments demonstrate that compared to methods such as mmDEAR and mmBaT, ActZ-Net reduces chamfer distance (CD) to 0.18, and increases point retention ratio (PRR) to 0.90 compared to mmDEAR and mmBaT, achieving 0.12 and 0.11 on Z-range error and Z-variance consistency (ZRE/ZVC) metrics, improving $Z$ -axis structural integrity and visual readability.

Zitieren

APA

Yin, W., Shi, L., & Shi, Y. (2026). Action-Semantics-Guided Z -Axis Point Cloud Reconstruction for Millimeter-Wave Radar. IEEE Transactions on Instrumentation and Measurement, 75, 8512111-8512111. https://doi.org/10.1109/tim.2026.3716503

Vancouver

Yin W, Shi L, Shi Y. Action-Semantics-Guided Z -Axis Point Cloud Reconstruction for Millimeter-Wave Radar. IEEE Transactions on Instrumentation and Measurement. 2026;75:8512111-8512111. doi:10.1109/tim.2026.3716503

BibTeX
@article{Yin2026,
  author = {Yin, Wei and Shi, Ling-Feng and Shi, Yifan},
  title = {Action-Semantics-Guided Z -Axis Point Cloud Reconstruction for Millimeter-Wave Radar},
  journal = {IEEE Transactions on Instrumentation and Measurement},
  year = {2026},
  volume = {75},
  pages = {8512111--8512111},
  doi = {10.1109/tim.2026.3716503}
}
Kostenlos starten →