Human Motion Analysis , Spring 2010

Human Motion Analysis

Spring 2010


This course will cover the theory and practice of human motion analysis using computer vision, machine learning and computer graphics techniques. We will expect students to have sufficient background to be able to read CVPR, NIPS and SIGGRAPH papers. We will review classic and contemporary methods for human motion classification, pose estimation and simulation. Representations of human motion, and classic human motion generation approaches including direct kinematics, inverse kinematics and motion grpahs will be reviewed at the beginning of the course. Discriminative approaches to tracking will be covered, including NN, regression techniques and Bayesian mixture of experts, as well as classic generative approaches to human body tracking such as the popular condensation algorithm, particle filters and likelihood models. Finally, we will review priors for human pose estimation and character animation including subspace models (e.g., PCA, GPLVM, Mixture of Factor Analyzers), joint limits and shape models. As time permits, we will cover related methods for gesture recognition based on human body motion, as well as physics-based approaches to tracking and character animation.

General information

Lecture: Mondays 2-5pm
Room: CHN D-48

Instructor: Raquel Urtasun
E -mail:


  • final project + oral exam.


  1. Introduction:
    1. what's going to be cover in the class?
    2. ways to capture human motion
    3. applications
  2. Human motion representations and direct kinematics:
    1. kinematic tree
    2. 3D locations
    3. joint angles
    4. axis angles
    5. quaternions
  3. Models of Human Pose and Motion:
    1. Latent variable models: PCA, FA, GPLVM, etc.
    2. Dynamical systems: LDS, etc
  4. Human motion synthesis:
    1. ML approaches: NN (motion graphs), Inverse kinematics, LVMs, etc.
    2. Space-time constraints: physics.
  5. Pose estimation from images:
    1. Inverse kinematics.
    2. Discriminative models: regression
    3. Generative models: Kalmann filters, particle filters, etc.
    4. Likelihood models.
    5. Priors: shape models, motion models, joint limits, physics.
  6. Human motion classification:
    1. Gait analysis
    2. Discriminative LVMs.
    3. Structured-output methods.


Lecture Date Topic


Readings Assignments
1 Feb 22 Introduction lecture1.pdf  


2 March 1 Human motion respresentations, direct kinematics and inverse kinematics
lecture2.pdf   exercises2.pdf (due March 8 at 2:00pm)
3 March 8 Dimensionality reduction I lecture3.pdf   No exercises. Colloquium talk by Bernhard Schölkopf (CAB G 51)
4 March 15 Dimensionality reduction II lecture4.pdf   exercises3_4.pdf (due March 29 at 2:00pm)

March 22

HMMs, Dynamical systems, Kalman filter lecture5.pdf   Continuation of exercises 3 and 4
6 March 29 Generative models: More on filters lecture6.pdf    
7 April 5, 12, 19 NO CLASS


8 April 26 Generative models: joint limits lecture7.pdf    
9 May 3 Generative models: shape priors, pose and motion priors lecture8.pdf    
10 May 6 Generative models: Likelihood models lecture9.pdf    
11 May 10 Generative models: physics lecture10.pdf    
12 May 17 Discriminative pose estimation lecture11.pdf    


14 May 31 Discriminative pose II. Combining generative and discriminative lecture12.pdf    


Research Projects

Link to the research projects available, and logistics.

The projects are due August 15. Send a report and code for the project. Presentations will take place the last week of August, same week as the exams.