Fachgebiet wearHEALTH

Winter semester 2021

We offer the Master level lecture Probabilistic methods for human motion modeling and capturing (KIS entry, OLAT link). Please book the OLAT course to register. For the course code, write an email to Bertram Taetz.

The kick-off event at 29.10.2020 will be online via BigBlueButton, starting at 12:00. For this, start the OLAT course and select Virtual classroom in the menu on the left side.

Summer semester 2021

We offer the Master level project and seminar Simulation, capturing and analysis of human motion

News

  • Registration closes at April 9, 2021.

Winter semester 2020/21

We offer the Master level lecture Methods for human motion modeling and capturing (KIS entry, OLAT link). Please book the OLAT course to register. For the course code, write an email to Gabriele Bleser.

The kick-off event at 30.10.2020 will be online via BigBlueButton, starting at 12:00. For this, start the OLAT course, select Virtual classroom in the menu on the left side, then select the session corresponding to the date.

Summer semester 2020

We offer the Master level project Simulation, capturing and analysis of human motion

The topic will be: Making human motion sound with neural networks

News

  • Registration is closed

Winter semester 2019/20

We offer the Master level lecture Methods for human motion modeling and capturing (KIS entry).

Fridays, 11:45 - 13:15, room 48-210

Description

In this lecture you learn how to deduce biomechanically interpretable movement data from sensor measurements (in particular body-worn inertial measurement units). The focus of the lecture will be on models and methods from sensor fusion (recursive and optimization based), system identification and machine learning. The needed basics in human anatomy will also be provided. In hands-on sessions you will get the opportunity to apply the learnt methods to real data captured in our motion lab.

The lecture addresses master level students of computer science and applied mathematics.

News

Material (to be linked here)

  • Lecture 1 (Introduction)
  • Lecture 2 (Human body - kinematic models - marker-based optical motion capture)
  • Lecture 3 (3D rigid body kinematics, motion lab visit)
  • Lecture 4 (Introduction to inertial motion capturing), Exercise sheet 1 (IMU data): session on December 6, 2019, 10-11:30, room 48-231, Solution 1
  • Lecture 5 (Introduction to Bayesian filtering)
  • Lecture 6 (Kalman filter), Exercise sheet 2 (Kalman filter): session on December 20, 2019, 10-11:30, room 48-231, Solution 2
  • Lecture 7 (Extended Kalman filter)
  • Lecture 8 (Application of the EKF to IMU based human kinematics estimation), Exercise sheet 3 (EKF based kinematics estimation and pose-based IMU-to-segment calibration): session on January 17, 2020, 10-11:30, room 48-231, Solution 3
  • Lecture 9 (Probabilistic sensor fusion and parameter estimation from an optimization perspective)
  • Lecture 10 (Bayesian linear regression)
  • Lecture 11 (Gaussian process regression)
  • Lecture 12 (Probabilistic deep neural networks)
  • Lab session (Probabilistic regression)

Summer semester 2019

We offer the Master level project  Simulation, capturing and analysis of human motion

News

  • The final presentations will take place at 21.08.19, 10-12 am, room 48-654
  • The intermediate meeting will take place at 28.05.19, 10-12 am, room 48-654
  • Please check the information on the seminar/project organisation here.

Topics

The available topics will focus on human motion simulation / behavior imitation learning

Example references:

  • Generative Adversarial Imitation Learning (GAIL): paper, code base
  • Learning human behaviors from motion capture by adversarial imitation: paper, code base

For students, who are interested in starting a master thesis in this area (e.g. after a seminar/project):

  • InfoGAIL: Interpretable Imitation Learning from Visual Demonstration: paper, code base

The idea is to apply the presented techniques to our own motion capture databases (walking of healthy people and patients) for automatically identifying different movement patterns (physiological gait, limping), for being able to sample motion patterns (data augmentation) or for matching to known gait patterns.

All code bases use Tensorflow for the machine learning part.

If you are interested, please write an email to Gabriele Bleser latest until 12.4.2019

Zum Seitenanfang