| |
| Research |
|
My research interests lie at the intersection
of computer vision and graphics. More specifically, I am
interested in
- Algorithms and techniques for recovering and
modeling the appearance, shape and motion
of real-world objects from video or multiple
views;
- Novel data-based representations:
local and image-based representations for rendering/visualization
of real-world objects;
- Methods for data and example-based
learning: extracting priors/patterns and probabilistic
models for shape/motion recovery and model acquisition.
My thesis
work focuses on probabilistic formulations for recovering
3D shape and motion from images and addresses the
issues of building theoretically sound frameworks
from first principles, learning strong priors and likelihood
distributions from example shapes and images, and the design
of probabilistic reconstruction algorithms.
|
|
|
|
|
|
|
| N-view stereo: Probabilistic
Inference of 3D Shape from Noisy Photographs |
|
| |
Abstract:
This work focuses on the inference of 3D shape
from a set of N noisy photos. We derive a
probabilistic framework to specify what one can infer about
3D shape for arbitrarily-shaped, Lambertian scenes and arbitrary
viewpoint configurations. Based on formal definitions of
visibility, occupancy, emptiness, and photo-consistency,
the theoretical development yields a formulation
of the Photo Hull Distribution, the tightest
probabilistic bound on the scene's true shape that can be
inferred from the photos. We show how to (1) express this
distribution in terms of image measurements, (2) represent
it compactly by assigning an occupancy probability
to each point in space, and (3) design a stochastic
reconstruction algorithm that draws fair samples (i.e., 3D
photo hulls) from it. |
|
Related publications:
- A
Probabilistic Theory of Occupancy and
Emptiness, Rahul Bhotika, David
J. Fleet, and Kiriakos N. Kutulakos, European
Conference on Computer Vision (ECCV 2002), Copenhagen,
Denmark, May 2002.
- A
Probabilistic Theory of Occupancy and
Emptiness: Detailed Analysis and Proofs,
Rahul Bhotika, David J. Fleet, and Kiriakos
N. Kutulakos,Technical Report 753, Computer Science
Department, University of Rochester, December 2001.
|
|
| Back to top |
|
|
|
| Model-based Tracking of
3D Non-Rigid Shapes in Video |
|
|
|
| Initial |
Refined |
| Model |
Model |
|
| 3D Tracking |
|
Abstract:
We have developed a linear framework for model-based
tracking of nonrigid 3D objects and for acquiring
such models from video. 3D motions and deformations
are calculated directly from image intensities without information-lossy
intermediate results. Measurement uncertainty is
quantified and fully propagated through the inverse
model to yield posterior mean and/or mode pose estimates.
A Bayesian framework manages uncertainty, accommodates priors,
and gives confidence measures. We obtained highly accurate
and robust closed-form motion estimators by
minimizing information loss from non-reversible (inner-product
and least-squares) operations, and, when unavoidable,
performing such operations with the appropriate error norm.
For model acquisition, we can refine a crude or generic model
to fit the video subject. Demonstrated uses are 3D tracking,
3D model refinement, and super-resolution texture
lifting from low-quality low-resolution video.
|
|
Related publications:
- Flexible
Flow for 3D Non-Rigid Tracking and Shape
Recovery, Matthew Brand and
Rahul Bhotika, IEEE Conference on Computer
Vision and Pattern Recognition (CVPR 2001), Kauai,
Hawaii, December 2001.
- Flexible
Flow for 3D Non-Rigid Tracking and Shape
Recovery, Matthew Brand and
Rahul Bhotika, TR 2001-38, Mitsubishi Electric
Resarch Labs (MERL), Cambridge,MA.
More
information
|
|
| Back to top |
|
|
|
Last edited July 25, 2002::Rahul Bhotika::
bhotikaATcs.rochester.edu
|