Three-dimensional Motion Capture, Modelling and Analysis of Ski Jumpers

1
Three-dimensional Motion Capture, Modelling and
Analysis of Ski Jumpers

• Atle NesCSGSC 2005 Trondheim, April 28th

2
Overview

1. Project description
2. What kind of data are we interested in?
3. Capturing data Image acquisition, Camera
   system
4. Processing data Feature points, Motion capture,
   Photogrammetry
5. Interpreting data Visualization, Motion
   analysis
6. Conclusion

3
Project description

• Task Design a computer system that can capture
  and study the motion of ski jumpers in 3D.
• Goal The results will be used to give feedback
  to the ski jumpers that can help them to increase
  their jumping lengths.

4
Data collection

• Will be gathered and analyzed in close
  cooperation with Human Movement Science Program
  at NTNU.
• Data
• Mainly from outdoor ski jumps captured at
  Granåsen ski jumping hill here in Trondheim.
• Also from indoor ski jumps captured at Dragvoll
  sports facilities.

5
Granåsen ski jump arena
6
Image acquisition

• Video sequences are captured simultanuously from
  multiple video cameras.
• Two decisive camera factors
• Spatial resolution (pixels)
• Time resolution (frame rate)

7
Camera equipment

• 3 x AVT Marlin F080b
• IEEE1394 Firewire, DCAM
• 8-bit greyscale w/ max resolution 1024x768x15fps
  or 640x480x30fps
• Extra trigger cable/signal ? Video capture
  synchronization.
• Different camera lenses ? Capture the same area
  from different distances.
• Optical fibre ? Extends the distance from
  computer to cameras in the hill, keeping the
  transmission speed.

8
Feature points

• Robust feature points
• Human body markers (easy detectable)
• Naturally robust features (more difficult).
• Want to have automatic detection of robust
  feature points using simple image processing
  techniques.

9
Motion capture

• Localizing, identifying and tracking identical
  feature points in both sequences of video images
  as well as accross different camera views.
• Synchronized video streams ensures good 3D
  coordinate accuracy.

10
Tracking w/ missing data
?

• Occluded features ? Redundancy using multiple
  cameras with different views.
• Probability theory ? Guess the point position
  based on feature point velocity.
• Another problem ? Blur effect

11
Photogrammetry

• Matching corresponding feature points from two or
  more cameras allows us to calculate the exact
  position of that feature point in 3D.
• Good camera placement is important for good
  triangulation capabilities (3D coordinate
  accuracy).

12
Camera calibration

• Coordinate system ? On site calibration using
  known coordinates in the ski jumping arena.
• Direct Linear Transformation (DLT) by Abdel-Aziz
  and Karara in 1971.
• Lens distortion (unlinear)
• Intelligent removal of the worst calibration
  points (sources of error).

13
Visualization

• Feature point tracks are connected back onto a
  dynamic model of the ski jumper.
• Dynamic model of ski jumper is combined with
  static model of ski jump arena.

14
Motion analysis

• Done in close cooperation with Human Movement
  Science Program
• Extract movements that have greatest influence on
  the result.
• Using statistical tools and prior knowledge about
  movements
• Project some movements to unseen 2D views.

15
Related applications

• Medical
• Diagnosis of infant spontaneous movements for
  early detection of possible brain damage
  (cerebral palsy).
• Diagnosis of adult movements (walk), for
  determination of cause of problems.

16
Related applications 2

• Sports
• Study top athletes for finding optimal movement
  patterns.
• Surveillance
• Crowd surveillance and identification of possible
  strange behaviour in a shopping mall or airport.

17
Conclusion

• I have presented an overview of a system that can
  capture, visualize and analyze ski jumpers in a
  ski jumping hill.
• Remains to see how well such a system can perform
  and if it can help the ski jumpers improve their
  skills.

18
Any questions?