Vision-Inspired Image Quality Assessment for Radar-Based Human Activity Representations

TL;DR

This work uses a benchmark radar dataset to reimplement and assess three recent denoising and preprocessing techniques: adaptive preprocessing, adaptive thresholding, and entropy-based denoising, and proposes a novel framework for static activity recognition using range-angle feature maps to expand HAR beyond dynamic activities.

Abstract

Radar-based human activity recognition has gained attention as a privacy-preserving alternative to vision and wearable sensors, especially in sensitive environments like long-term care facilities. Micro-Doppler spectrograms derived from FMCW radar signals are central to recognizing dynamic activities, but their effectiveness is limited by noise and clutter. In this work, we use a benchmark radar dataset to reimplement and assess three recent denoising and preprocessing techniques: adaptive preprocessing, adaptive thresholding, and entropy-based denoising. To illustrate the shortcomings of conventional metrics in low-SNR regimes, we evaluate performance using both perceptual image quality measures and standard error-based metrics. We additionally propose a novel framework for static activity recognition using range-angle feature maps to expand HAR beyond dynamic activities. We present two important contributions: a temporal tracking algorithm to enforce consistency and a no-reference quality scoring algorithm to assess RA-map fidelity. According to experimental findings, our suggested techniques enhance classification performance and interpretability for both dynamic and static activities, opening the door for more reliable radar-based HAR systems.

Figures (8)

• Figure 1: Spectrogram after adding 5 different White Gaussian noise levels.
• Figure 2: Spectrogram results for Walking (W): original spectrogram (top-left), noisy spectrogram (top-right) and 4 methods under WGN (SNR = 10 dB).
• Figure 3: Spectrogram results for Walking (W): original spectrogram (top-left), noisy spectrogram (top-right) and 4 methods under WGN (SNR = -5 dB).
• Figure 4: Our measurement setup in the long-term care facility (left) and our laboratory (right).
• Figure 5: Grad‑CAM visualization on Range–Azimuth feature maps for frame $5$ of static activities. Top row: class A0 (lying on floor); bottom row: class A2 (sitting on sofa). In each row, the left panel shows the raw RA map with the red box indicating the ground‑truth target region, the center panel overlays the Grad‑CAM heatmap (magma colormap), and the right panel highlights the network’s attended region (green box).