Explore the evolution and challenges of estimating human motion in this comprehensive lecture by Michael Black. Delve into the historical context, key concepts, and technological advancements in the field of human motion estimation. Learn about various techniques such as relaxation labelling, kinematic trees, stochastic search, and graphical models. Discover the complexities of modeling the human body, including shape identity, scanning methods, and deformation processes. Gain insights into the development of virtual humans, the importance of body shape and pose estimation, and the application of these concepts in tracking infants and analyzing interpersonal interactions. Understand the progression from simple body models to more complex representations, and explore the potential future directions in this fascinating area of research.
Estimating Human Motion: Past, Present, and Future
Overview
Syllabus
Intro
What Im interested in
Goals
Why is it hard
History of the field
Posttalk
Relaxation Labelling
Computer Vision
Kinematic Tree
The Lost Decade
The Basic Paradigm
Sandy Pentland
Dario Gavrila
The Human Body
Stochastic Search
Simulated annealing
Particle filtering
Priors
Priors are a crutch
Graphical models
Early graphical models
Higher dimensional models
The basic idea
The problem
Cyberware scanner
Why are bodies hard
The Cesar dataset
The Skate Model
The First Paper
Virtual Humans
Scanning the Human Body
Collecting Existing Data
Professional Models
Mesh
Shape Identity
Scanning
Body Shape
Poses
Template Mesh
Shape Blend
Pose
Linear Blend Skinning
Pose Blend Shapes
Deformation
RGBD
Track Infants
Interpersonal Interaction
Simple Body Model
Simple X
Taught by
Andreas Geiger
