man M otion S imilarity Francisco J Torres Reyes Outline of the Talk Problem and Challenges 3D ChainCode LABANotation Contribution 1 Comparison Analysis of ChainCode and FastDTW ID: 632137
Download Presentation The PPT/PDF document "MUMS a M easure of h U" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
MUMSa Measure of hUman Motion Similarity
Francisco J Torres ReyesSlide2
Outline of the TalkProblem and Challenges3D ChainCodeLABANotationContribution 1: Comparison Analysis of ChainCode and FastDTW
Contribution 2: Enhanced
LABANotation
for RehabilitationContribution 3: System Architecture for HMTRLessons LearnedFuture DirectionConclusion
2
4/24/14
Ftorres/MUMSSlide3
Why Measuring Human Motion Similarity?34/24/14Ftorres/MUMSSlide4
How is Human Motion Captured and Modeled?Sports – high speed camera for slow motion speed analysisVR – real time data acquisition -> simulation timeModeling human body -> skeleton44/24/14
Ftorres/MUMSSlide5
Human Motion can be modeled as Sets of 3D Curves3D Curves shown by MUMS tool from a shoulder exercise. I developed this tool for Windows environmentThese two diagrams capture the accumulated tracks of two snapshots.It shows all sensors data including those from head, torso, abdomen, arms and legs.5
4/24/14
Ftorres/MUMSSlide6
Utilize Time3D data from HMTR ProjectFour Key Rehabilitation Exercises were chosen and analyzed in HMTR project by Dr. Yunyu Wang – Certified Movement Analysis, LABANotation Reconstructor and Teacher
Dr. James
Carollo
– Physical Medicine and Rehabilitation, Ortophaedics, Bioengineering, School of Engineering and Applied Science.6
4/24/14
Ftorres/MUMSSlide7
C3D File ContentMoCap Data# Pts = 34# Video Frames = 1238 Video Frame # 1
1 733.4042 1447.329 1659.807
2 720.8265 1322.164 1659.62
… 33 452.7758 1343.254 38.04591
34 666.9445 1325.737 36.31926
Video Frame #
2
…
7
3D data per marker
Sampling rate
Body position for markers
4/24/14
Ftorres/MUMSSlide8
Chain Code - Orthogonal Changes of DirectionInvariant under Translation or Rotation[Bribiesca, 2006]8
4/24/14
Ftorres/MUMSSlide9
Measuring similarity on chain codes9Slide10
Measuring similarity on chain codes10Slide11
Adding Time into the EquationThe United States National Anthem
What
is an equivalent representation for human motion?11
4/24/14
Ftorres/MUMSSlide12
LABANotationLABANotation: a record of how one moves so it can be repeated. This notation includes a set of symbols that are placed on a vertical staff, where its vertical dimension represents the symmetry of the body, and its horizontal one represents the time[Bouchard, 2008]12Slide13
Spatial and Temporal Analysis134/24/14Ftorres/MUMSMeasures of 3 beats display from bottom up. Different movements of body limbs are encoded with directions
Movement on the right leg track can be encoded in
chain code for motion analysis.
Note that there are five sensors per leg.Therefore five corresponding chain code may be generated.Slide14
Dynamic Time WarpingFinds the optimal alignment between two time seriesUse the value calculated based on the optimal alignment to represent the similarity.If two time series are the same, the similarity value is zero.14
4/24/14
Ftorres/MUMS
Y2X4
X2
X
2
X
4
Y
2
d
(Y2, X4)+d(Y3,X5)+d(Y4,X6)+d(Y5,X7)+d(Y6,X8)=
0+0+0+0+0=0
>
d(Y2,
X2)
+d(Y3,
X3)
+d(Y4,
X4)
+d(Y5,
X5)
+d(Y6,
X5)=1+1+0+1+0+1=
4
Similarity values contributed by
subsequence pairs:Slide15
Slow Start vs. Fast Pace – 3D ChainCode15
Slow start
Fast pace
Idle at starting position
Idle at ending position
New Idle symbol
same 3D curvesSlide16
Slow Start rotated 90o – 3D ChainCode and FastDTW16
4/24/14
Ftorres/MUMS
Slide17
Slow Start rotated 270o – 3D ChainCode and FastDTW17
4/24/14
Ftorres/MUMS
Slide18
Shoulder Elevation and Rotation Exercise18Similarity values of arms time3D curves are close with 3D ChainCode Similarity Formula
Similarity values of arms
time3D
curves are quite different with
fastDTW resultsSlide19
Standing Hip Abduction Exercise194/24/14Ftorres/MUMSSlide20
Mini Squat Exercise204/24/14Ftorres/MUMSSlide21
Contribution 2: Enhanced LABANotation for RehabilitationLABANotation is designed to describe dance.We studied its usage and suggested the enhancement for rehabilitation purposes.Focus on the Add movement precision by adding new symbolsMinimize notation modifications and changesApply the new notation on improving the specification of key rehab exercises in the HMTR project.4/24/14Ftorres/MUMS21Slide22
Enhanced LABANotation for Rehabilitation – Mini Squats ExerciseStart standing with equal weight distributed between right and left legsPlace feet shoulder width apartKeep torso upright, avoid bending at the waistSlowly loser yourself by bending ankles, knees, and hipsReturn to standing22
4/24/14
Ftorres/MUMSSlide23
Enhanced LABANotation for Rehabilitation – Standing Hip Abduction ExerciseStart standing with equal weight distributed between right and left legsSlowly, shift your weight to the left sideRaise the leg out to the side ~ 12’’Keep the right foot facing forwardKeep the torso upright
and avoid leaning
to the side23
4/24/14
Ftorres/MUMSSlide24
Contribution 3: HMTR System Design and Tool EvaluationWe propose a Human Motion Tracking and Reasoning (HMTR) Software Architecture.Evaluates Tools for HMTR System DesignLabanWriter (Mac version from Ohio State)LabanDancer
(Windows version from Dance Bureau)
4/24/14Ftorres/MUMS24Slide25
Proposed Human Motion Tracking and Reasoning (HMTR) Software Architecture254/24/14Ftorres/MUMSSlide26
Suggested Enhancement for LabanDancer Software[Wilke, Calvert, Ryman, 2005]264/24/14Ftorres/MUMSSlide27
Lessons LearnedC3D data is a binary data difficult to parse. Use tool from Internet to extract into text form and feed them in chain code program.The original 3D ChainCode dissimilarity algorithm is very slow when applying to real exercise data. The steps are re-examined and improved for the time performance. Bribiesca’s group did not consider the idle situation and did not encode the elapsed of time. They are interested in shape of the curves, while we are interested in movement. Laban Dancer executable code works fine but the source code was compiled in different version of visual studio and even the original authors can not provide a working project.
27
4/24/14
Ftorres/MUMSSlide28
284/24/14Ftorres/MUMSUse this framework in another areas, i.e. dancing, video gamesImprove run time for the analysis of human motion, i.e. using nVidia CUDA toolsGetting additional data with semantics for further analysis of similarityDefine additional properties such off-track, sustain and develop procedure for computing the values for the properties.
Future ResearchSlide29
ConclusionProposed a new model to represent human motionUsed LABANotation to analyze human motion on spatial and temporal domainsSuggested enhancement of LABANotation for rehabilitationDeveloped a software tool to perform the analysis of human motion similarity on motion capture sessions
Proposed an HMTR
software architecture
Propose enhancements for LabanDancer software for rehabilitation
purposesThe
analysis
of human motion is needed in different areas of study.
Three papers were published. Will submit the work on the comparison of chain code and
fastDTW
.
29
4/24/14
Ftorres/MUMSSlide30
304/24/14Ftorres/MUMSSlide31
314/24/14Ftorres/MUMS