Human Motion Information

1
Human Motion Information

• Device Development and Application to Motion
  Analysis

by Jijun Wang
2
Introduction

• Motivation
• The nature of human movement
• People with disability
• Human factor
• Robot control
• Other

3
Introduction (Cont.)

• Object
• Developing methods to capture information about
  human motion
• applying them to human motion analysis
• Content
• 2D motion information detection
• 3D motion information detection
• movement analysis for index finger target-reaching

4
2D Motion Information Detection

• Method
• Attach landmarks to human body
• Use video camera to record human bodys movement
• Apply software to extract human motion information

• Key Technology
• Extracting motion information from video images

5
2D Motion Information Detection (Cont.)

• Motion Information Extraction
• Image Pre-process
• Object Eliminate noise and distortion from video
  images
• Method Edge Detection Approach that is based on
  fuzzy sets
• Description
• Image Matrix
• Enhanced Image
• Function ,

6
2D Motion Information Detection (Cont.)

• example

Original Image
Processed Image
7
2D Motion Information Detection (Cont.)

• Image Recognition
• Object Distinguish landmarks
• Method An algorithm based on continuous area
  recognition
• Description
• Two kinds of search
• Find all landmarks (continuous areas with
  specified features) in the whole video image
• Predict the position of a landmark, then seek it
  in the predicted area

8
2D Motion Information Detection (Cont.)

• The motion analysis software

9
3D Motion Information Detection

• Method
• With two or more cameras, we can get an objects
  3D coordinates from the images taken by the
  cameras.

10
3D Motion Information Detection (Cont.)

• DLT (Direct Linear Transformation) algorithm
• Lets say
• (X,Y,Z) is the objects 3D coordinates
• (a,ß) is the image coordinates of the object
• F is the relationship function between (X,Y,Z)
  and (a,ß). That is (a,ß) F (X,Y,Z)
• Calibration
• For a known set of (X,Y,Z,a,ß), we can get the
  relationship function F. That is (X,Y,Z,a,ß) gt
  F
• Reconstruction
• For a known set of (a,ß) and the relationship
  function F, we can get the 3D coordinates. That
  is (X,Y,Z) F-1(a,ß)

11
3D Motion Information Detection (Cont.)

• Error analysis of DLT
• Error sources
• The algorithm
• The parameters obtained from calibration
• The 2D digitized coordinates obtained from video
  images
• The 2D digitized coordinates
• The errors of 2D digitized coordinates cant be
  ignored
• Researchers have seldom studied its influence on
  calibration

12
3D Motion Information Detection (Cont.)

• The influence of image coordinates on calibration
• Object
• How does the image coordinate error affect
  reconstruction?
• How to limit the influence?
• Method
• Simulate the error of image coordinates
• Apply calibration procedure to the image
  coordinate
• Investigate the reconstruction error

13
3D Motion Information Detection (Cont.)

• The Numeric Experiment

Reconstruction errors vs. intersection angle
between cameras
Dx0.005m for image coordinates
14
3D Motion Information Detection (Cont.)

• Conclusion
• in 3-D measurement, the space constituted by all
  the reference points should wrap the potential
  measured space
• the reference points should be evenly and
  uniformly distributed in the potential measured
  space
• The 3D measure system
• The calibration frame

15
3D Motion Information Detection (Cont.)

• An example

The left image
The right image
16
Index Fingers Target-reaching Movement

• Introduction
• The human finger is the most precise human
  instrument
• Target-reaching movement is a very complex
  movement with multiple degree of freedom
• Benefits
• Evaluate artificial finger
• Control theory
• Robot control (Inverse Kinematics)

17
Index Fingers Target-reaching Movement (Cont.)

• Motivation
• How to assess the fingers movement?
• Are there some motion patterns to describe finger
  movement?
• Method
• Attach landmarks to index finger
• Use 3D measurement device to collect finger
  movement information
• Analysis of movement data

18
Index Fingers Target-reaching Movement (Cont.)

• The Experiment

19
Index Fingers Target-reaching Movement (Cont.)

• Motion quality evaluation
• Index of difficulty and performance
• 1D translational movement (Fitts Law) (P. M.
  Fitts, 1954)
• Id (Index of difficulty)
• Ip (Index of performance)
• 1D angular movement (Extended Fitts Law) (G. V.
  Kondraske, 1995)
• Id? (Index of difficulty)
• Ip? (Index of performance)

20
Index Fingers Target-reaching Movement (Cont.)

• 2D Movement (Extended Fitts Law)

• The acceleration time constant
• The velocity index the average speed throughout
  a task
• The time index the time spent on the task

21
Index Fingers Target-reaching Movement (Cont.)

• The power index the total energy (ET) consumed
  in the whole movement
• The smoothness index the average instantaneous
  smoothness index Smoothness throughout a task
• Other accessory indices
• the covariance of speed the maximal
  instantaneous speed the maximal instantaneous
  kinetic energy the covariance of smoothness the
  maximal instantaneous smoothness.

22
Index Fingers Target-reaching Movement (Cont.)

• Examples

A high level difficulty
B normal level difficulty
C low level difficulty
23
Index Fingers Target-reaching Movement (Cont.)

• Conclusion
• The experimental results show that these indices
  could describe the point-touch movement of index
  finger
• Which index should be used for evaluation depends
  on the knowledge about the movement or the
  statistical data

24
Index Fingers Target-reaching Movement (Cont.)

• Preliminary research on motion pattern
• Object
• Try to find the general motion pattern of
  fingers target-reaching movement
• Method
• Normalize target-reaching movement
• Use Principal Component Analysis (PCA) to find
  factors that have significant effect on finger
  movement

25
Index Fingers Target-reaching Movement (Cont.)

• Conclusion
• The topological invariance of human movements
  exists among different subjects and different
  tasks
• Its affected by the start position and end
  position of the movement
• The PCA result also shows there is a power index
  that affects the motion pattern

The normalized target-reaching trace

• Further study

26

• Thank you