COMETH: Convex Optimization for Multiview Estimation and Tracking of Humans

Check out our paper and project page.

In the era of Industry 5.0, monitoring human activity is essential for ensuring both ergonomic safety and overall well-being. We propose COMETH (Convex Optimization for Multiview Estimation and Tracking of Humans), a lightweight algorithm for real-time multi-view human pose fusion that relies on three concepts:

• it integrates kinematic and biomechanical constraints to increase the joint positioning accuracy
• it employs convex optimization-based inverse kinematics for spatial fusion
• it implements a state observer to improve temporal consistency.

The proposed fusion pipeline enables accurate and scalable human motion tracking, making it well-suited for industrial and safety-critical applications.

Installation

Create the virtual environment (runnning Python 3.7):

python3.7 -m virtualenv .venv
source .venv/bin/activate

Install the dependences:

pip install -m requirements.txt

Create the COMETH package:

pip install -e .

COMETH Applications

Multiview Skeleton Fusion

A novel convex optimization-based framework for real-time multi-view human pose fusion.

Example

To run the complete aggregator example, please refer to this repository:

• sync_dynamic_aggregator.py

Compared Methods

• Boldo et al. 2024
• OpenPTrack

Reference

@article{Martini2026,
  title = {COMETH: Convex optimization for multiview estimation and tracking of humans},
  volume = {314},
  ISSN = {0957-4174},
  url = {http://dx.doi.org/10.1016/j.eswa.2026.131728},
  DOI = {10.1016/j.eswa.2026.131728},
  journal = {Expert Systems with Applications},
  publisher = {Elsevier BV},
  author = {Martini,  Enrico and Choi,  Ho Jin and Figueroa,  Nadia and Bombieri,  Nicola},
  year = {2026},
  month = jun,
  pages = {131728}
}

IMU-HPE Sensor Fusion

A sparse sensor-fusion framework for upper-limb pose estimation with shoulder-mounted IMUs and a single chest-mounted egocentric camera.

Example

A complete example can be found in the Jupyter Notebook IMU-HPE_fusion/test_our_model.ipynb.

Compared Methods

• Li2022 implementation: IMU-HPE_fusion/test_li2022_model.ipynb
• EKF (ZeroVel) implementation: IMU-HPE_fusion/test_ekf_zv_model.ipynb

Reference

Currently under review, stay tuned!