Thomas Pollard

PhD Candidate
Division of Applied Mathematics
Brown University
182 George St.
Providence, RI 02912, USA
Thomas_Pollard AT brown DOT edu

Research

My research is in computer vision, specifically problems in change detection. The primary goal of my work is to develop a next generation 3-d change detection algorithm which works under arbitrary camera and lighting configurations. The framework I've developed uses voxels which model both surface likelyhood and appearance, so abnormal appearances in a new image with any viewpoint can be detected as change. The first phase of this research involved detecting changes in a 3d scene taken from arbitrarily positioned cameras such as these helicopter images of Providence:

Some samples from a 34 image sequence used to learn the normal background appearance.

Detected changes in a new image with an unseen viewpoint.

A volume rendering of the recovered most-probable surface voxels.

This work is explained in detail in our 2007 CVPR paper: Change Detection in a 3-d World

Current research activities include extending this algorithm to work under arbitrary but known lighting conditions. Recently I've also been applying my work to other related problems such as image based rendering and height-map estimation from aerial imagery.

Advisor: Joseph L. Mundy

Curriculum Vitæ

Updated: 12/18/2007