From Neck to Head: Bio-Impedance Sensing for Head Pose Estimation

TL;DR

This work introduces NeckSense, a camera-free head pose estimator that leverages multi-channel bio-impedance sensing around the neck. A five-electrode neckband collects impedance magnitude and phase, which are fed into Imp2Head, a transformer-based encoder–decoder that maps temporal impedance features to SMPL-X head/neck/jaw rotations while enforcing biomechanical constraints. Evaluation on seven participants against a vision-based ground truth shows strong performance, with MPJPE/MPVE improving from baseline to 6.7 mm/5.9 mm when biomechanical priors are included, and a corrected MPJPE of 25.9 mm compared to OS-X, demonstrating competitive viability with state-of-the-art vision methods. The approach offers a practical, LOS-free alternative for robust head pose tracking in AR/VR and HCI, with potential for long-term wearability due to the soft, dry electrode necklace design. Future work includes personalization, auto-calibration to mitigate placement drift, and evaluation with true MoCap ground truth.

Abstract

We present NeckSense, a novel wearable system for head pose tracking that leverages multi-channel bio-impedance sensing with soft, dry electrodes embedded in a lightweight, necklace-style form factor. NeckSense captures dynamic changes in tissue impedance around the neck, which are modulated by head rotations and subtle muscle activations. To robustly estimate head pose, we propose a deep learning framework that integrates anatomical priors, including joint constraints and natural head rotation ranges, into the loss function design. We validate NeckSense on 7 participants using the current SOTA pose estimation model as ground truth. Our system achieves a mean per-vertex error of 25.9 mm across various head movements with a leave-one-person-out cross-validation method, demonstrating that a compact, line-of-sight-free bio-impedance wearable can deliver head-tracking performance comparable to SOTA vision-based methods.

Figures (4)

• Figure 1: Bio-impedance-based head pose estimation uses five necklace-integrated electrodes to monitor neck impedance changes caused by head movement, which depend on tissue composition, electrode geometry and spacing, and muscle orientation or structural deformation
• Figure 2: NeckSense uses five soft, reusable electrodes around the neck, with one for stimulation and four for impedance sensing (magnitude and phase). Signals from the AD5941 are read by an ESP32-S2 and sent via Bluetooth for real-time processing
• Figure 3: Architecture of the Imp2Head model, a modular encoder-decoder transformer that maps temporal impedance signals to joint rotations. The encoder captures temporal patterns, while the decoder predicts joint angles autoregressively using attention over encoded features. The loss combines mean squared error with biomechanical constraints to ensure anatomically valid predictions.
• Figure 4: Head pose estimation from neck impedance signals: The left column displays time-series neck bio-impedance signals, the middle column shows the predicted SMPL-X poses derived from these signals using Imp2Head, and the right column provides reference ground-truth images of the subjects in corresponding poses.