Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative Latent Alignment for Interpretable Radar Based Occupancy Detection in Ambient Assisted Living

Huy Trinh

TL;DR

This work tackles privacy-preserving, interpretable occupancy detection using mmWave radar in Ambient Assisted Living. It introduces Generative Latent Alignment (GLA), which combines a lightweight convolutional VAE on Range–Angle heatmaps with a frozen CLIP-based text-anchoring head and latent Grad-CAM to generate semantically grounded explanations. The latent space is softly aligned to two anchors, 'empty' and 'person present', yielding class-conditional Grad-CAM maps that localize presence-related RA structure; ablations show radar-specific anchors are crucial for meaningful explanations. The approach offers privacy-preserving, interpretable radar-based occupancy detection with improved transparency for clinicians and operators in AAL settings.

Abstract

In this work, we study how to make mmWave radar presence detection more interpretable for Ambient Assisted Living (AAL) settings, where camera-based sensing raises privacy concerns. We propose a Generative Latent Alignment (GLA) framework that combines a lightweight convolutional variational autoencoder with a frozen CLIP text encoder to learn a low-dimensional latent representation of radar Range-Angle (RA) heatmaps. The latent space is softly aligned with two semantic anchors corresponding to "empty room" and "person present", and Grad-CAM is applied in this aligned latent space to visualize which spatial regions support each presence decision. On our mmWave radar dataset, we qualitatively observe that the "person present" class produces compact Grad-CAM blobs that coincide with strong RA returns, whereas "empty room" samples yield diffuse or no evidence. We also conduct an ablation study using unrelated text prompts, which degrades both reconstruction and localization, suggesting that radar-specific anchors are important for meaningful explanations in this setting.

Generative Latent Alignment for Interpretable Radar Based Occupancy Detection in Ambient Assisted Living

TL;DR

This work tackles privacy-preserving, interpretable occupancy detection using mmWave radar in Ambient Assisted Living. It introduces Generative Latent Alignment (GLA), which combines a lightweight convolutional VAE on Range–Angle heatmaps with a frozen CLIP-based text-anchoring head and latent Grad-CAM to generate semantically grounded explanations. The latent space is softly aligned to two anchors, 'empty' and 'person present', yielding class-conditional Grad-CAM maps that localize presence-related RA structure; ablations show radar-specific anchors are crucial for meaningful explanations. The approach offers privacy-preserving, interpretable radar-based occupancy detection with improved transparency for clinicians and operators in AAL settings.

Abstract

In this work, we study how to make mmWave radar presence detection more interpretable for Ambient Assisted Living (AAL) settings, where camera-based sensing raises privacy concerns. We propose a Generative Latent Alignment (GLA) framework that combines a lightweight convolutional variational autoencoder with a frozen CLIP text encoder to learn a low-dimensional latent representation of radar Range-Angle (RA) heatmaps. The latent space is softly aligned with two semantic anchors corresponding to "empty room" and "person present", and Grad-CAM is applied in this aligned latent space to visualize which spatial regions support each presence decision. On our mmWave radar dataset, we qualitatively observe that the "person present" class produces compact Grad-CAM blobs that coincide with strong RA returns, whereas "empty room" samples yield diffuse or no evidence. We also conduct an ablation study using unrelated text prompts, which degrades both reconstruction and localization, suggesting that radar-specific anchors are important for meaningful explanations in this setting.
Paper Structure (6 sections, 8 equations, 3 figures)

This paper contains 6 sections, 8 equations, 3 figures.

Table of Contents

  1. Related Work
  2. Methodology
  3. Semantic Radar VAE
  4. Text-Anchor Latent Alignment with Frozen CLIP
  5. Latent-Space Explainability (Latent Grad-CAM)
  6. Experiments and Results

Figures (3)

  • Figure 1: The proposed Semantic Radar VAE architecture. The model learns to reconstruct the input Range-Angle heatmap while simultaneously aligning its latent representation with frozen CLIP text anchors for "Empty" and "Person" classes. This enables Latent Grad-CAM visualization of class evidence.
  • Figure 2: Qualitative interpretability for an Empty-room RA sample using the text-aligned VAE and latent Grad-CAM.
  • Figure 3: Qualitative interpretability for a Person-present RA sample. The latent Grad-CAM localizes the presence-related RA blob.