Augmented parametric shapes for real-time dense 3D modeling using an RGB-D camera

Friday, December 9, 2016,11:00 am to 12:00 pm, room F108, INRIA Montbonnot

Seminar by Diego Thomas, University of Kyushu, Fukuoka, Japan


Abstract: Consumer grade RGB-D cameras such as the Kinect camera or the Asus Xtion pro camera have become the commodity tool to build dense 3D models of indoor scenes. The volumetric Truncated Signed Distance Function (TSDF) and its extensions have become popular and largely used for the task of 3D modeling using RGB-D cameras. However, this representation suffers several limitations due to its important memory footprint and unpractical manipulations. We propose a novel compact, flexible and accurate 3D representation based on parametric surfaces augmented by 2D deviation images. We demonstrate the usefulness of our proposed 3D representation with applications to real-time large-scale indoor scene reconstruction and dynamic 3D face modeling. First, we propose a method to build large scale indoor scenes by fitting planar patches with deviation images to the input data. While the set of planar patches provides a structured representation of the scene useful for loop closure processes, the deviation images encode the fine 3D details of each object in the scene. Second, we propose a method to build in real-time animated 3D head models using a consumer-grade RGB-D camera.The facial motion is captured using a blendshape animation model while the geometric details are captured with a deviation image mapped over the template mesh. We demonstrate robust and high-fidelity simultaneous facial motion capture and 3D head modeling results on a wide range of subjects with various head poses and facial expressions.

Bio: Diego Thomas received his master’s degree in Informatics and Applied Mathematics from the Ecole Nationale Superieure d’Informatique et de Mathematiques Appliquees de Grenoble (ENSIMAG), France in 2008. He received his Ph.D. from the Graduate University for Advanced Studies (SOKENDAI), Japan in 2012. He was a post-doc researcher at the National Institute of Informatics (Tokyo, Japan) from April 2012 to March 2015. Since April 2015, he is a JSPS post-doc researcher at the University of Kyushu (Fukuoka, Japan). His research interests are 3D images registration, 3D reconstruction and photometric analysis.