3D Human Pose Estimation (Technology Focus)

Reconstruct human posture and poses quickly and accurately in 3D from a single RGB image.

Leveraging state-of-the-art object-detection, novel human mesh models and neural network architecture this 3D Human Pose Estimation technology reconstructs human poses from a single RGB image eliminating the need for complex multi-camera setups, wearable markers, expensive scanners and the need to calibrate them. 

Achieve a much higher level of performance(?) for industrial robot simulations and human-robot collaboration, trajectory prediction, collision avoidance, AR/MR applications, medical training systems, sports visualization systems.


The POC tests show state-of-the-art accuracy in mesh reconstruction and faster inference than previous methods, particularly in multi-person environments.


Accurate (mean per joint error of 66 mm on POC dataset)


Faster inference (2.2 FPS vs. 0.7 FPS of current state of the art methods)

Simple Setup

No expensive sensor setups. Only requires a commercial RGB camera


Standalone (no external person detectors)

3D Skeleton and Mesh Reconstrcution

Detects person bounding box, reconstructs 3D skeleton and 3D mesh

Compatible with Rendering Engines

Versatile output → Compatibility (Unity, Blender, Maya …)
The output of the system is a set of parametric human meshes that are compatible with rendering software (Blender, Maya) and simulation and game engines such as Unity. 

Use Cases

The developed POC shows the advantages of this system for applications that require real-time human shape and pose reconstruction. Some interesting use cases for the final model could include:

  • Augmented and Mixed reality: integrating real and simulated/virtual environments with people. Uses in telepresence, medical training systems, sports visualization systems.
  • Industrial robot simulations: human-robot collaboration, industrial safety, human obstacle avoidance (robot trajectory planning).
  • Human motion or trajectory prediction, action recognition: this system can be coupled with further developments for augmented functionalities.

Get Started with Human Pose Estimation