Files

OneShotCalibration_TIM2026 (17).pdf
  • Open Access
  • Adobe PDF
  • 22.48 MB
Download

Details

Authors
Abstract
Radars offer robust gesture recognition capabilities over traditional vision, capacitance, and ultrasound sensors, with flexibility in preserving user privacy, sensing in different lighting conditions, higher temporal resolution, and sensing through materials. However, raw radar signals are cluttered, difficult to interpret , and strongly dependent on the radar hardware and scene configuration, which often forces gesture recognizers to be retrained for each new radar and environment. To address these limitations, we build on a full-wave electromagnetic radar model and inversion-based approach to calibrate and normalize radar measurements into a physically meaningful Green's function that is independent of the radar hardware and background scene. In this paper, we (1) experimentally validate the radar system invariance of the model by comparing signals from two distinct radars observing the same scene and frequency range, demonstrating that their diverse raw signals align to each other after normalization; (2) analyze the sensitivity of the processing pipeline to the distance and angle of the target, revealing the physical limitations of the infinite-plane assumption for finite hands; and (3) empirically evaluate the method on a comprehensive dataset of 20 gestures in user-dependent, user-independent, and mixed scenarios. Our results achieve classification accuracies of 90.58%, 97.78%, and 98.54% for the full set, and subsets of 12 and 8 gestures, respectively, confirming the approach's robustness and flexibility.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Arthur Sluyters, Lambot, S., Attygalle, N., & Vanderdonckt, J. (2026). An Electromagnetic Model and Inversion-based Approach for Radar Gesture Recognition. IEEE Transactions on Instrumentation and Measurement, 76. https://doi.org/10.1109/TIM.2026.3729019 (Original work published 2026)