Framework for 3D Motion Field Estimation and Reconstruction

We propose the framework for the motion estimation of 3D objects based on the motion vectors that form motion fields and the motion reconstruction based on the 3D rotations and the factorization method. The classical two-dimensional motion field approach is extended to three dimensions, e.g. on the volumetric objects that are moving in time. When dealing with the real world multiple moving objects and the complex scenes, lots of objects are moving with different motions, both in space and in time. In this context, an object can be described as a part of a scene that moves with a coherent motion and the scene can be broken down into a number of regions, each of which can be well approximated by its own motion. The first part of this paper describes the motion vectors based techniques which are used for the motion estimation, and the second part addresses problems with the reconstruction of the 3D motion and structure. Proposed methods estimate reconstruction from a sparse set of the matched volume features on the 3D neuroimage in NIfTI format. We evaluate these techniques according to the different problems they address.