Statistical and Variational Methods for Problems in Visual Control

Abstract

Following the position of moving objects based on the information delivered by single or multiple optical sensors (e.g., video cameras) is the objective of visual tracking. The need for visual tracking is ubiquitous and a multitude of approaches exist for the solution of this tracking problem. Increasingly, computer vision algorithms are required to provide additional information beyond a simple track point, and more complex methodologies are needed to produce the desired information. For noisy, cluttered, and/or dynamic scenes the ability to provide a smooth and faithful signal is essential which leads of course to the entire issue of filtering. In this research program, we have developed a novel visual tracking approach, using statistical variational methods. In particular, we have developed a geometric particle filter for controlled active vision. This has been applied to various tracking problems including tracking through turbulence and UAVs flying in formation.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 02, 2009
Accession Number
ADA531631

Entities

People

  • Allen Tannenbaum

Organizations

  • Georgia Tech

Tags

Communities of Interest

  • Biomedical
  • Materials and Manufacturing Processes
  • Space
  • Weapons Technologies

DTIC Thesaurus Topics

  • Algorithms
  • Calculus Of Variations
  • Computational Science
  • Computations
  • Computer Science
  • Computer Vision
  • Computers
  • Detectors
  • Diagnostic Imaging
  • Differential Equations
  • Filtration
  • Image Processing
  • Image Segmentation
  • Partial Differential Equations
  • Sequential Monte Carlo Methods
  • Unmanned Aerial Vehicles
  • Variational Methods

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Computer Vision.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms