detection algorithm of throwing form changing to prevent ...tetujin/pervasive2012/dalt-slide.pdf ·...
TRANSCRIPT
DALT : Detection Algorithm of Throwing Form
Changing to Prevent the Baseball Players’ Throwing Related Injuries
Keio UniversityYuuki Nishiyama
12年6月10日日曜日
Abstract• We achieved creating the algorithm (DALT)
to detect the change in throwing form by using wearable gyroscope sensor.
12年6月10日日曜日
Introduction• In baseball, the injury of shoulder and
elbow(throwing injury) issues for many baseball players
- it is caused by overwork or unsuitwork- 72% of high school baseball players experience
the injury in Japan
• the player has to manage their condition by themselves
- there are fewer numbers of coaches than baseball players in a team
12年6月10日日曜日
Status of Throwing Injury-Questionnaire of the Keio Univ. and high school baseball team-
• Questionnaire survey (to 181 players)[Fig.1] 57% of players have experience of throwing injury
[Fig.2] 63% of players practicing with throwing injury even now
[Fig3.] 83% of throwing injuries are the cause of overwork
43%57%
have an experience don’t have an experience
20%
38%43%
usuallysometimesseldom or never
13%17%
33%
37%
overworkbad formeitheranother
Fig1. experience of throwing injury
Fig2. throwing injury even now
Fig3. cause of throwing injuries
12年6月10日日曜日
Throwing motion• Throwing motion is consist of five
phases(Frank Jobe)- Wind-up, Early-cocking, Late-cocking,
Acceleration, Follow-through
• Definition of Motion Points- Takeback, Cock, Top, Maxout, Release- between phase
and phaseTakeback Cock Top Maxout Release
Wind-up Early-cocking Late-cocking Acceleration Follow-through
12年6月10日日曜日
Pathogenic mechanism of throwing injuries
• Fatigue of deltoid muscle and supraspinatus muscle from increase in number of throw
• Increase of shearing strength to shoulder joint at late-cocking
• Accumulation of fatigue to joint lip and a ligament
• Lead to throwing injure - SLAP lesion,Bennett lesion
Takeback Cock Top Maxout Release
Wind-up Early-cocking Late-cocking Acceleration Follow-through
12年6月10日日曜日
Method of prevention of throwing injuries• Limit the number of throwing
- Limit of the number of the throws in Little League‣ http://www.littleleague.org/media/newsarchive/2007stories/
07pitching_rule_reg_changes_10_07.htm
• Detection fatigue statue- blood lactate level- measurement of joint motion and power- analysis throwing from
12年6月10日日曜日
Related work
• High speed camera (DLT)• 3D motion capture system• Motion sensor
(acceleration and gyro sensor)
12年6月10日日曜日
Problem
• The perception of objective throw fatigue is difficult by self management
• Existing systems have difficulty in use by the exercise
12年6月10日日曜日
Approach
• Detect the throw form change with increase of the number of he throws
• Using only one sensor device- 3D wireless gyro sensor
12年6月10日日曜日
DALT:Detection Algorithm of Throwing Form Changing
• Sensor was put on the lower part of upper arm
- 200Hz, 1500deg/sec(LOGICAL PRODUCT)
• Using personal throw data( =base throwing data )
• Three step of DLAT1. Detection of the Throwing Motion2. Decomposition of Throwing Form3. Calculate Rate of Throwing Form Change
Z
Y X
12年6月10日日曜日
Angular velocity data of throwing
-400-300-200-1000
100200300400500
ZYX
Takeback Cock Top Maxout Release
12年6月10日日曜日
Base Throwing Data1. The personal data of angle velocity in
throw motion- Normal throw from ×10 throws- Fatigue throw from ×10 throws
2. Detect the release ; I will state this later3. Save the data for 400 frames from release4. Calculate data of average from ten throws
12年6月10日日曜日
1. Detection of the Throwing Motion
1. Save 400 frame datas(angle verosity)2. Detect of release
- using threshold3. Judge of throwing motion
- using correlation coefficient
STEP1 STEP2 STEP312年6月10日日曜日
• Y-axis value becomes 0 at near release- “Measurement of motion of upper limb and trunk with inertial
sensors during baseball pitch”
2011, Kenji saito, Journal of Nagoya Gakuin University
• A shape upper arm internal rotation motion and external rotation
- Increase of shape Y-axis value
• Threshold...1. Y-axis : -350deg/sec2. Y-axis : 0deg/sec
‣ release point
Detect of release
-400-300-200-1000
100200300400500
Time ( 1/200sec )
deg/sec
Y-angleX-angle
Z-angle
1. Y < -350deg/sec
2. Y = 0deg/sec = release
-350deg/sec
12年6月10日日曜日
Judge of throwing motion using coefficient correlation• Calculate a coefficient correlation of
NormalData and NowData• Judge the throw motion from result of
coefficient correlation- Y-axis > 0.8- X-axis > 0.5- Z-axis > 0.5
STEP1 STEP2 STEP312年6月10日日曜日
2. Decomposition of Throwing Form• Decomposition throwing form from Model
of kinetic chain- back calculation from release- detect distinctive actions
in order of release, maxout, top, cock, takeback
STEP1 STEP2 STEP3
motion points distinctive data
release start position of the analysis
maxout Y-axis < 0top Z-axis > 0cock Y-axis > 0
takeback X-axis > 0
12年6月10日日曜日
3. Calculate Rate of Throwing Form Change
• The acquisition of angular velocity data between Top and Maxout
- Now form data (nowData)- Normal form data (normalData)- Fatigue form data (fatigueData)
• Calculate a coefficient correlation(NowR) of nowData and normalData
• Calculate a coefficient correlation(BaseR) of normalData and fatigueData
• Rate of throwing form change(Rtc) = Now / BaseR×100
STEP1 STEP2 STEP3
angular velocity(deg/sec)
time( 1/200sec )normal form datafatigue form data
Maxout Release
12年6月10日日曜日
Hardware organization• Program language
- java
• PC of data analysis- MacBook Pro 15inch- 8GB, Intel Core 2 Duo
• Using sensor- 9 axis wireless sensor
‣ Logical Product
SpecificationsAngular velocity sensor ±1500dps
Sampling interval 200Hz
Size 55mm×40mm×22mm
Weight 35g
12年6月10日日曜日
Software configuration
Data of angular velocity
Rate of throwing form change
Personal throwing data of angular velocity
Module of decomposition of throwing form
Module of calculating rate of throwing form change
Part of coefficient correlation
Part of detecting release
Save 400 frame datas
Module of detection of throwing motion
The layer of calculating change of throw motion
Layer of judging throw motion
基準角速度データ
基準角速度データ
angular velocity
angular velocity
angular velocity
angular velocity angular velocity
angular velocity
angular velocity
12年6月10日日曜日
Evaluation Experiment• Goal of evaluation
- Evaluation of suggestion algorithm (DALT)
• Method of evaluation- 100 pitches of consecutive throws of rundown (27.431m)
‣ Ability of the ball control
‣ DTL-method : 3D motion analysis using two high speed camera
‣ DALT : Detection Algorithm of Throwing Form Changing
• Examinee- College baseball player × 5
• Place - The Keio university’s
baseball field
Z
Y X
12年6月10日日曜日
Estimation Procedure1. Divide result of 100 throws into three terms
- 1~10(Term1), 45~55(Term2), 91~100(Term3)2. Analysis of variance (ANOVA) at each three
terms3. Analysis of Dunnett’s test as the multiple
comparisons- term1 is control term
12年6月10日日曜日
Result of ANOVA• Ball control ability had no significant
- ball control ability could not be a barometer of fatigue state
• DALT and DLT-method were different significantly
- DALT and DLT-method could be a barometer of fatigue state
Methods Average (±Standard Deviation)Average (±Standard Deviation)Average (±Standard Deviation) F p SignificanceMethodsTerm1 Term2 Term3
F p Significance
Ball control Ability 3.44(±1.24) 3.56(±1.36) 3.3(±1.28) 0.23 0.8 n.s.
DTL-method 11.41(±3.95) 12.43(±4.48) 15.36(±5.99) 8.7928 0.0002 **
DALT 40.71(±22.21) 70.64(±29.9) 73.04(±32.12) 20.38 0.0001 **
F : F-distribution * : p<0.05 ** : p<0.001 n.s. : not significantF : F-distribution * : p<0.05 ** : p<0.001 n.s. : not significantF : F-distribution * : p<0.05 ** : p<0.001 n.s. : not significantF : F-distribution * : p<0.05 ** : p<0.001 n.s. : not significantF : F-distribution * : p<0.05 ** : p<0.001 n.s. : not significantF : F-distribution * : p<0.05 ** : p<0.001 n.s. : not significantF : F-distribution * : p<0.05 ** : p<0.001 n.s. : not significant12年6月10日日曜日
Evaluation of Dunnett’s test• DALT was significantly different between
Term1 and Term2, and Term1 and Term3• DTL-method was significantly different
between Term1 and Term 3
0
20
40
60
80
100
1 2 3
Rtc
(%)
Group
****
****
0
5
10
15
20
25
1 2 3
Max
-SA
(ang
le)
Group
**n.s
**n.s
n.s : not significant * : p < 0.05 ** : p < 0.00112年6月10日日曜日
Result of Analysis Time
• DLT-method- 45minutes / 1 throw
• DALT- 22.5 seconds / 1 throw
12年6月10日日曜日
Conclusion
• We proposed the Detection algorithm of throwing form changing(DALT)
- DALT can detect throwing form changing- DALT analyzes state in mach shorter time
than DAT-method- DALT is executed automatically
12年6月10日日曜日
Future work• The implementation of the learning
system• Necessary to conduct the experiment
with various distance • Comparison with the 3D motion capture
system• Real time feedback system
12年6月10日日曜日
Injures of shoulder joint
• SPLA lesion- Superior labrum anterior and posterior lesion
• Bennett Lesion• Shoulder Impingement Syndrome• Dead arm syndrome
12年6月10日日曜日
SLAP lesion-superior labrum anterior and posterior lesion-
• Superior Labrum Anterior and Posterior
• Term definition- articular lip- long head of biceps brachii muscle- infraspinatus muscle- deltoid muscle
12年6月10日日曜日