Stochastic Tracking of Humans

Keynote talk, Vision Interface 2000
Montreal, Canada

05/17/2000


Click here to start


Table of Contents

Stochastic Tracking of Humans

Goal: 3D Human Motion

Overview

Collaborators

Why is it important?

Why is it hard?

Why is it hard?

Why is it hard?

Other Problems

Common Assumptions

Requirements

Bayesian Inference

Problems

Bayesian Formulation

Generative Model: Shape

Generative Model: Appearance

Appearance Model

Noise Model

Generative Model: Temporal

Bayesian Formulation

Robust Likelihood

Likelihood

Temporal Model: Smooth Motion

What does the posterior look like?

Particle Filtering

Representing the Posterior

Stochastic Search

Condensation

Condensation

Condensation

Condensation

Visualizing Results

Arm Tracking: Smooth motion prior

Full-Body Tracking

Learning Temporal Models

Detecting Cycles

Modeling Cyclic Motion

Modeling Cyclic Motion

Action-Specific Model

Temporal Model: Walking

Learned Walking Model

Learned Walking Model

Learned Walking Model

Learned Walking Model

Learned Walking Model

Stochastic 3D Tracking

Stochastic 3D Tracking

No likelihood

Issues

Lessons Learned

Work to be done

Outlook

Related Work

Conclusions

Author: Michael J. Black

Email: black@cs.brown.edu

Home Page: http://www.cs.brown.edu/people/black/