Construction of 3D statistical deformable models for deformable objects with applications in object reconstruction and motion recognition


Student thesis: Doctoral Thesis

View graph of relations


  • Peng DU

Related Research Unit(s)


Awarding Institution
Award date3 Oct 2012


With the development of 3D scanner and capture devices and computer modeling tools, deformable 3D objects have become easy to obtain and have been used in a wide spectrum of fields. How to statistically modeling the deformable 3D objects for various applications has become an active research topic. This thesis provides sophisticated ways to construct 3D Statistical Deformable Models (SDMs) for two different forms of deformable 3D objects, i.e., 3D surface form and 3D skeleton form, for object reconstruction and motion recognition, respectively. For the 3D surface form objects, the small sample size problem is frequently encountered when constructing SDM for them, due to their high data dimensions. To address this problem, this thesis proposes to construct piecewise SDM based on divide-and-conquer strategy instead of single global SDM for the objects. To construct a piecewise SDM, two key steps are required: partitioning the surface into multiple components and assembling the local deformed SDMs to form the final SDM for the object surface. Studying on two different kinds of 3D surface data, we propose respective techniques for constructing piecewise SDMs for them. The medical 3D surface data derived from CT images has a special multi-layer structure that is relatively easier to process. We propose to construct a hierarchical piecewise SDM consisting of a coarse global SDM built on feature points of the surface and a set of local SDMs. The global SDM provides a framework for partitioning the surface, and also for assembling local SDMs. On the other hand, the generic 3D surface data is much more difficult to deal with. For the surface partitioning problem, we propose to partition a surface according to the similarity of the surface variability characteristics, and subsequently propose two novel measures for quantifying the variability similarity. For the assembly problem, we propose a technique based on constrained deformation for seamlessly stitching the local deformed SDMs. The piecewise SDMs for the two kinds of 3D surface data are both applied to object reconstruction. For ensuring the global shape consistency of the entire piecewise SDM, we further propose a multi-level SDM based technique to constrain the deformation of the local SDMs. For the 3D skeleton form objects, e.g. the motion capture data, we propose to construct a behavior-specific SDM for each type of the motions to capture the common characteristics and the possible deformation modes shared by them. The ability of SDM to encode all allowable deformation for the object class it represents makes the behavior-specific SDMs quite suitable for motion classification. To make use of this advantage, a new motion is classified based on how well each behavior-specific SDM represents it, i.e., how accurately each behavior-specific SDM can reconstruct the new motion. We show this novel technique is more powerful for motion classification compared with the traditional eigen-motion based classification technique.

    Research areas

  • Deformation potential, Motion perception (Vision), Three-dimensional imaging, Computer vision, Computer simulation, Image reconstruction