Millimeter-wave Imaging for Anthropometric Body Measurement

Miriam Senne, Benjamin D. Killeen, Christoph Baur, Nassir Navab, Azade FarshadarXiv 2026 · arXiv

TL;DR

This paper introduces a privacy-preserving, clothing-tolerant framework for extracting clinical body measurements using millimeter-wave (mmWave) radar imaging. By fitting a parametric human body model (SMPL) to noisy, partial radar point clouds using a novel vertex-weighting strategy, the system accurately captures 3D body shapes and circumferences without requiring patients to undress.

What Problem Does This Paper Address?

Anthropometric measurements—such as waist, hip, and chest circumferences—are vital clinical biomarkers used to evaluate health risks like cardiovascular disease and obesity. However, traditional data collection methods suffer from major bottlenecks:

  • Manual tape measurements are highly operator-dependent and prone to positioning errors.
  • Optical 3D scanners provide precision but require patients to undress and maintain specific poses.

These requirements slow down clinical workflows, compromise patient dignity, and effectively exclude vulnerable populations, such as elderly individuals or patients with limited mobility.

To bridge this gap, the authors explore millimeter-wave (mmWave) radar. While mmWave signals can penetrate clothing and preserve privacy by avoiding RGB cameras, they generate highly incomplete and noisy 3D point clouds due to sensor self-occlusion and multipath reflections. This paper addresses the challenge of accurately reconstructing a complete 3D human mesh from these imperfect radar scans.

Methods

The authors propose an optimization-based framework that registers a parametric human body model (SMPL) directly to a point cloud extracted from front and back mmWave scanner panels.

Fig. 2: Method Overview. The MI volume is converted to a point cloud and rigidly aligned with an initialized SMPL model in A-pose. Per-vertex weights are computed from the posed SMPL normals with respect to the panel normals. The distance between SMPL vertices and point cloud is then minimized to obtain the final shape parameters β used for anthropometric measurements.

The core technical contributions include:

  1. Dynamic Vertex Weighting (DVW): Standard Chamfer distance optimization struggles with partial point clouds because it penalizes missing data regions. The authors introduce a weighting strategy based on the squared cosine similarity between the SMPL mesh's surface normals and the scanner panel's orientation. Vertices facing away from the scanner panels receive lower weights, preventing unobserved regions from distorting the fit.
  2. Foot-Ground Plane Constraint: To prevent vertical drifting caused by weak radar returns from the feet, an energy term penalizes the distance between the foot soles of the model and the ground height.
  3. ISO-Standardized Extraction: Once the model is fitted, the system extracts circumferences by intersecting the mesh with horizontal planes at specific anatomical heights, computing a convex hull to simulate an actual physical measuring tape according to ISO 8559 standards.

Datasets

The method was evaluated using two distinct setups:

  • Mannequin Benchmark: A rigid physical mannequin was used to establish a baseline. Ground truth was captured via a high-resolution Artec Leo handheld 3D laser scanner. Measurements were taken under multiple conditions: unclothed, dressed in tight thermal wear, and wearing a loose blazer.
  • Real Human Dataset: A cohort of 27 human subjects (18 male, 9 female) spanning a wide BMI range ($16.64$ to $37.53 \text{ kg/m}^2$). Each participant was scanned twice: once in regular daily clothing and once in tight-fitting clothing. Expert manual tape measurements served as the human ground truth.

Key Insights and Ideas

  • Physics-Informed Loss Modulation: Instead of trying to "clean" the radar data beforehand, incorporating the sensor's line-of-sight geometry directly into the optimization loss function (via vertex weights) allows the algorithm to naturally ignore noise and missing data.
  • The Fallibility of "Ground Truth": The authors demonstrate that even on a rigid mannequin, manual tape measurements vary by up to 0.5 cm between trials, and even tight clothing adds over 1 cm to circumferences. This underscores the need for digital, hands-off measurement tools to ensure consistency across longitudinal clinical studies.
  • True Clothing Tolerance: Because mmWave radar bypasses outer garments, the framework achieves similar measurement accuracy whether a subject is wearing a heavy blazer or tight clothing, a feat impossible with standard optical or camera-based methods.

Key Results

The framework was benchmarked against a state-of-the-art monocular RGB baseline (SHAPY).

  • Mannequin Evaluation: The mmWave-based measurements closely matched the gold-standard 3D laser scan. Even when the mannequin wore a blazer, the calculated waist circumference was identical to the unclothed 3D laser scan baseline (81.0 cm).
  • Human Evaluation: The mmWave approach significantly outperformed the RGB baseline in Mean Absolute Deviation (MAD) relative to manual tape measurements. For example, the RGB baseline showed a chest circumference error of nearly 7 cm under loose clothing, whereas the neutral SMPL model on mmWave data maintained a MAD of 3.01 cm.
Fig. 3: Real data. Example mmWave point clouds and SMPL fits for two par- ticipants, with mmWave- and tape-based anthropometric measurements (right).
  • Consistency Across Clothing States: The difference in extracted measurements between clothed and unclothed states was remarkably low for the mmWave method (e.g., a delta of just 0.96 cm for the waist), proving its robustness against everyday attire.

Limitations and Future Work

Despite its strong performance, the authors note a few limitations:

  • Under-sampled Flanks: Because the scanner utilizes discrete front and back panels, the lateral sides (flanks) of the body and the head suffer from sparse data coverage, which can occasionally lead to minor alignment errors.
  • Parametric Model Constraints: Relying strictly on statistical body models like SMPL or STAR means the system inherits the shape priors of those datasets, which may underrepresent highly unique or non-standard body topologies.

Future work will focus on relaxing the parametric constraints to recover highly precise, subject-specific body surfaces and expanding benchmarks against a wider variety of alternative 3D imaging technologies.

Summary updated

Abstract

Body shape and circumferences are clinically informative biomarkers for risk stratification, including measures such as waist to hip ratio, limb and trunk girths, yet conventional tools such as manual tape measures and optical scanners often require undressing and sustained poses. These demands slow workflows, compromise dignity, and exclude many older adults and people with limited mobility. To make measurement fast and contactless, we leverage millimeter-wave (mmWave) radar, which preserves privacy and operates through typical clothing, enabling quick full-body acquisition. In this work, we present a new optimization-based framework to recover 3D human shape and extract a comprehensive set of anthropometric measurements from volumetric mmWave data. Our method introduces a weighted registration pipeline that fits a parametric body model (SMPL) directly to the noisy mmWave point cloud. The core of our contribution is a vertex-weighting strategy that modulates a Chamfer energy function for reliable surface alignment and noise elimination. We further stabilize the fit by incorporating a foot-ground plane constraint and pose priors, optimizing directly for the SMPL parameters. Together, these components enable a fast, privacy preserving workflow that delivers high fidelity body shape and measurements through clothing without cameras or disrobing and with minimal cooperation, supporting frequent risk oriented assessments in clinics and care facilities for patients of all ages and mobility levels.
BibTeX
@misc{senne2026millimeter,
  author = {Miriam Senne and Benjamin D. Killeen and Christoph Baur and Nassir Navab and Azade Farshad},
  title = {{Millimeter-wave Imaging for Anthropometric Body Measurement}},
  year = {2026},
  eprint = {2605.23064},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2605.23064},
}

Loading related content...

Figure