About

Project SHARED - Shape Analysis and Registration of People Using Dynamic Data

Project SHARED seeks to investigate novel shape analysis methods that exploit large redundancy of information from dynamic or movement data.

First, a large enough collection of dynamic, time-varying data sets should be acquired. Appreciably, with the recent advances in imaging technologies we now have growing accessibility to capture shape and motion of people with high frequency. For example, we track the location of selected skin surfaces (markers) using a number of synchronized cameras. After post-processing of marker trajectories we obtain sequences of human body motions in the form of animated meshes.

Second, a shape model should be constructed. The model should be faithful to dynamic properties of deforming surfaces that are obtained through data acquisition

Third, we want to move towards devising new non-rigid registration and spatio-temporal segmentation techniques. We have to put entirely new perspective on the traditional registration pipeline, trying to understand where and how the mobility can be encapsulated. The dynamic features along with shape-based geometric features will be used to guide the full correspondence as well as the transformation for the optimal match.

The SHARED objectives could be briefly described as follows:

Design shape acquisition and analysis method where kinematic properties can be fully learned; Develop new registration techniques that uses the kinematic properties obtained above so as to put similar deformation behaviours in correspondence; Develop statistical model that incorporates the dynamic shape variation and shape identity variation; Revise the registration technique with the statistical shape atlas, towards a more robust and efficient application.