Body Surface Reconstruction from a Small Set of Samples


Background:

The purpose of this project is to reconstruct the body surface of a Motion Capture (MoCap) performer from a small set of samples taken before the performance. We already have such a tool in our lab, but it constitutes a low resolution, crude approximation of the body surface which looks more like a “block-man” (blue character on the image above). We want to exploit the a priori knowledge about the curvature structure of the human body and obtain a proper looking humanoid.

Outcomes:

The primary goal is to deform a given mesh to fit it to the body surface of a person by using a set of positional samples and simple body measurements of the performer (height, limb lengths etc.) and save it into an industry standard format. The student will also be responsible for deforming the underlying skeleton to be able to animate the character afterwards. According to the progress other goals could defined:

  • Indexing the resulting mesh in the form of a lower resolution mesh to access its elements efficiently
  • Adding extra deformations to the resulting mesh (adding a fat belly etc.) from a simple UI.

Development Framework:

The development of this project will be on the top of our state of the art user calibration framework which is written in C++. The output mesh will be saved in any of the formats that Unity3D Game Engine can import automatically (.FBX, collada, .obj…). Therefore, the student needs to learn an API for saving the resulting mesh in one of these formats.

The student will also require to implement some functionalities within Unity3D Game Engine in C#.

Requirements:

Programming (C++ for mesh deformation, C# for Unity3D)

Contact:

Eray Molla (eray.molla@epfl.ch, INJ139)

Ronan Boulic (ronan.boulic@epfl.ch, INJ141)