Jonathan W. Brandt

Visiting Associate Professor

Research Interests

3D Motion and Structure from Image Sequences

The goal of this project is to automate as much as possible the construction of computer graphics scene models from raw, uncalibrated video sequences. That is, starting with a video clip of a static scene that has been digitized, process this clip off-line in order to generate a 3D description of the scene in the form of texture-mapped polygonal surfaces. This description can later be displayed interactively on a graphics workstation. The surfaces are recovered through image motion analysis and the textures are obtained from the digitized video images.

Accurate Measurement of Optical Flow

Optical flow refers to the instantaneous velocity of brightness patterns in time-varying images. The optical flow field contains potentially useful information about the 3D motion and structure of the imaged scene. In order to make use of this information, it is vital to understand and quantify the various sources of error in optical flow estimation. Systematic study of such errors leads to more accurate estimation methods, and also a deeper understanding of the fundamental limitations to accurate optical flow estimation.

Object-Oriented Software for Computer Vision

The specification and development of reusable objects for computer vision tends to simplify and accelerate the creation of large-scale vision systems. In the process, new understanding of the fundamental structures in computer vision is achieved.

Prototype

Photo Album

Resume

Publications

Bookmarks