GetMobile • 2025
MoCaPose: Motion Capture with Textile-Integrated Capacitive Sensors – A New Approach to Wearable Tracking
Highlights
- Decouples sensor placement from anatomical alignment using capacitive sensing and deep regression
- Loose-fitting smart garments with textile-integrated electrodes and low-power CDCs
- Empirical mapping from multi-channel capacitance to 3D joint positions without explicit physics models
- Dataset: 38 hours, 21 participants covering diverse body types and motions
- Accuracy comparable to IMU systems: MPJPE ~86 mm; R-squared up to 0.932
- Performance degrades with faster motions; static poses more accurate
- Generalizes across users; leave-person-out MPJPE varied by less than 5%
- Sensors can be repositioned (e.g., ±5 cm at shoulders) without recalibration
- Pose features boost unsupervised HAR, outperforming raw-signal baselines by 34–37%
- Open-source hardware, data, and ML pipelines for community use
Abstract
This article presents MoCaPose, a wearable motion capture approach that integrates multi-channel capacitive sensors into loose-fitting garments and uses deep learning to regress continuous 3D human poses. The method decouples sensor placement from anatomical joints by learning the latent relationship between body geometry and capacitive signals, avoiding explicit physics-based modeling. Two textile prototypes were developed, culminating in a robust mesh-jacket design with integrated conductive traces and FDC2214 capacitance-to-digital converters. Trained on 38 hours of data from 21 participants, MoCaPose achieves IMU-comparable accuracy (MPJPE around 86 mm) and strong pose correlation (R-squared up to 0.932), remaining robust across body types and sensor repositioning. Converting signals to pose representations improves downstream activity recognition, and the project releases open-source designs, datasets, and pipelines.
Click to read Full Paper.