EUVIMEDEUVIMED entdecken
Crossref

Adaptive radar–thermal fusion for robust human motion recognition in dynamic environment

Qinwei Li, Bangxuan Rao, Cheng Fang, Hang Wu, Ming Yu

Measurement Science and Technology · 13.11.2026 · 37 (45) · 455110

DOI: 10.1088/1361-6501/aeb4e0 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 13.11.2026. Verfasst von Qinwei Li und 4 weiteren Autor:innen. Die Arbeit nennt 21 Literaturangaben. Gefördert von: Tianjin Education Commission.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study investigates subject-independent radar-thermal human motion recognition under modality-specific environmental interference. A synchronized dual-modal sensing system was developed using an IWR6843ISK millimeter-wave radar and a 24 × 32 thermal infrared array. A multi-participant dataset containing 720 ten-second synchronized radar-thermal sequences was collected from ten participants performing five actions—standing, walking, waving, jumping, and squatting—under clean, thermal-hot-object, and radar-moving-reflector conditions. The proposed framework constructs a motion-enhanced multimodal representation by combining radar point-cloud descriptors, thermal-distribution descriptors, sequence-level motion descriptors, invariant temporal-spectral characteristics, and radar-thermal interaction information. A subject-independent multi-model fusion (SIMMF) mechanism then integrates complementary probability predictions generated from multiple representation views and classification models. An auxiliary subject-disjoint adaptive probability refinement (SDAPR) strategy is further evaluated as a conservative post-refinement stage. Generalization to unseen participants is assessed using a strict nested leave-one-subject-out protocol, in which the outer-test participant is excluded from all feature standardization, model selection, probability fusion, calibration, and fallback decisions. SIMMF correctly classified 647 of the 720 outer-test sequences, corresponding to a pooled accuracy of 89.86% and a macro-F1 score of 89.88%. SDAPR corrected three additional SIMMF errors, increasing the final pooled accuracy to 90.28% and the macro-F1 score to 90.30%; an exact two-sided McNemar test yielded p = 0.25. The corresponding final accuracies under the clean, radar-moving-reflector, and thermal-hot-object conditions were 94.58%, 88.75%, and 87.50%, respectively. These results support recognition within the tested cohort and configurations, with SDAPR providing only a marginal auxiliary refinement. Broader interference robustness and superiority over external studies remain unestablished.

Förderung

  • Tianjin Education Commission

Zitieren

APA

Li, Q., Rao, B., Fang, C., Wu, H., & Yu, M. (2026). Adaptive radar–thermal fusion for robust human motion recognition in dynamic environment. Measurement Science and Technology, 37(45), 455110. https://doi.org/10.1088/1361-6501/aeb4e0

Vancouver

Li Q, Rao B, Fang C, Wu H, Yu M. Adaptive radar–thermal fusion for robust human motion recognition in dynamic environment. Measurement Science and Technology. 2026;37(45):455110. doi:10.1088/1361-6501/aeb4e0

BibTeX
@article{Li2026,
  author = {Li, Qinwei and Rao, Bangxuan and Fang, Cheng and Wu, Hang and Yu, Ming},
  title = {Adaptive radar–thermal fusion for robust human motion recognition in dynamic environment},
  journal = {Measurement Science and Technology},
  year = {2026},
  volume = {37},
  number = {45},
  pages = {455110},
  doi = {10.1088/1361-6501/aeb4e0}
}
Kostenlos starten →