Kinematic 3D Geometry Modeling of Captured Human Characters

Realistic animation of human characters is required for many applications, e.g. movie and game productions. Building on top of our previous work [1,2,3], we develop here a geometric 3D representation to model human characters, which is highly realistic as well as efficient and simple to animate. For realistic modeling of articulated motion for a wide range of poses, a combination of Linear Blend Skinning (LBS) and Dual Unit Quaternion Skinning (DQS) is used. This non-linear skinning approach nicely compensates for the deficiencies (Candy-wrapper, bulging artifacts etc.) of both methods by decomposing the joint rotations. DQS is used for rotations around the outward bone axis while LBS models bending.

This model consists of a set of vertices, skinning weights, joint locations and joint orientations. Further, we develop an optimization framework to adapt all these components to a set of registered example 3D scans of an actor in different poses. This optimization loops of the 3 following steps until convergence:

  • optimize skinning weights via coordinate descent,
  • optimize vertices via least squares fitting,
  • optimize joint locations and orientations via Gauss-Newton algorithm.

The objective function to minimize consists of a data term, enforcing that transformed model vertices are close to their targets, and a weighted Laplace prior in order to enforce smoothness among neighboring vertices. The weight of the Laplace prior is the only parameter required to be specified.


P. Fechteler, W. Paier, A. Hilsmann, P. Eisert:
Real-time Avatar Animation with Dynamic Face Texturing, Proceedings of the 23rd International Conference on Image Processing, Phoenix, Arizona, USA, 25th - 28th September 2016.

P. Fechteler, A. Hilsmann, P. Eisert:
Example-based Body Model Optimization and Skinning, Eurographics 2016 Short Paper, Lisbon Portugal, May 2016.

P. Fechteler, W. Paier, and P. Eisert:
Articulated 3D Model Tracking with on-the-fly Texturing, Proceedings of the 21st International Conference on Image Processing (ICIP), Paris, France, 27th - 30th October 2014.

P. Fechteler, A. Hilsmann, and P. Eisert:
Kinematic ICP for Articulated Template Fitting, Proceedings of the 17th International Workshop on Vision, Modeling and Visualization (VMV), Magdeburg, Germany, 12th - 14th November 2012.

P. Fechteler and P. Eisert:
Recovering Articulated Pose of 3D Point Clouds, Proceedings of the 8th European Conference on Visual Media Production (CVMP), London, UK, 16th - 17th November 2011.