Extrac'ngTennisSta's'cs fromWirelessSensing Environments - PowerPoint PPT Presentation

Extrac'ng
Tennis
Sta's'cs
 from
Wireless
Sensing
 Environments
 Kenneth
Conroy
(CLARITY),
Adel
Shaeib
&
Mark
Roantree
 (ISG)
 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Overview
 Collabora'on
with
Tennis
Ireland
 Forms
part
of
a
complete
tennis
coaching
system
(TennisSense)
 U'lises
wireless
sensing
environment
(Ubisense)
 A
process
of
enrichment
and
court
mapping
allows
querying
through
XPath
 &
XQuery
 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Ubisense
Setup
 Ubisense
sensors
located
in
6
points
around
court
 Indoor
localisa;on
 Players
hold
Ubitags
in
their
pockets
 Tracked
by
Ubisense
system
 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Architecture
 P1
takes
raw
data
and
converts
to
XML
 P2
seman'cally
enriches
the
data
using
domain
specific
knowledge
 P3
transforms
high‐level
queries
into
lower
level
queries
based
on
states
 and
zones.
 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Enrichment
 XML 
 Raw 2 <match
players="2"
dura'on="27">
 kevin <player
name="kevin"
id="0"
entries="6881">
 eric <ac'on
'mestamp="15.625">
 15.625 <x>‐0.467</x>
 0 <y>4.575</y>
 -0.467003226280212 <z>1.199</z>
 4.57561922073364 <side>lee</side>
 1.19958519935608 <change_side>0</change_side>
 15.625 
 <game>1</game>
 <zone>lee_baseline_bogom</zone>
 <special_zone>lee_bogom_special
 </special_zone>
 </ac'on>
 </player>
 Raw
text
converted
to
XML
file
 XML
provides
structure,
seman'c
enrichment
is
also
performed
(game,side
etc)
 Domain
specific,
can
be
applied
to
addi'onal
sports
without
significant
changes
to
 the
overall
structure
 Each
ac'on
can
be
easily
accessed
and
interpreted

 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Court
layout
 Zones
consist
of
(x,y)
range
values
corresponding
to
court
posi'ons
 Zones
provide
structure
from
which
queries
can
be
constructed
 Queries
transformed
from
high‐level
to
low‐level
using
zonal
informa'on
&
enriched
XML
 Detect
serve
involves
both
players
being
in
a
par'cular
posi'on
for
a
defined
'me
period
 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Queries
 XML
format
allows
detailed
queries
for
Tennis
coaches
(Number
of
serves
in
 game
1,
Posi'on
of
player
1
when
player
0
was
in
a
special_zone)
 XQuery
Example
1:
How
many
data
values
(Ubisense)
did
player
1
record
 during
the
match?
 let
$a
:=
doc("leName.xml")/match/player[@id="1“]
return
string($a/ @entries)
 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Performance
 Detect Serve Performance: The system can detect 74.7% of serves, and an additional 18.3% are detected when 2-detected- as-1 detections are included Enrichment is very quick, even for very large files as is the time taken to perform queries UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE



Summary
 U'lise
XML
to
provide
structure
to
raw
data
 Define
zones
on
the
court
&
Iden'fy
structure
and
rules
of
tennis
 Use
structure
and
rules
to
define
queries
 Provide
an
interface
for
the
coach
to
query
 High
level
of
accuracy
at
this
early
stage
 Far
less
computa'onally
expensive,
and
more
cost
effec've
than
alternate
 approaches
 Current
work
is
focussing
on
velocity/accelera'on
and
using
addi'onal
 sources
of
informa'on
to
improve
accuracy
 UNIVERSITY
COLLEGE
DUBLIN


  


DUBLIN
CITY
UNIVERSITY


  


TYNDALL
NATIONAL
INSTITUTE