受賞論文ダウンロード PDF 526 K

)HDWXUH$UWLFOH
0DVDR+RULED$ZDUGV
Technical Reports
堀場雅夫賞特別賞受賞者論文
Ubiquitous Geo-Localization Without Infrastructure
インフラを使用しないユビキタス位置特定手法
The primary method for geolocalization is based on GPS which has issues of
Alper YILMAZ
localization accuracy and unavailability. We introduced a new geolocalization
アルパーイルマズ
approach to GPS-denied indoor and outdoor environments. His approach has
two principal components: public domain transport network data for outdoor
and indoor available in GIS (Geographic Information System) databases; and
trajectory data acquired from a mobile platform. This trajectory is estimated using
inertial sensors or visual odometry. We abstract the transport map information
as a graph data structure, where various types of roads are modeled as graph
edges and typically intersections are modeled as graph nodes. A search for the
trajectory in real time in the graph yields the geolocation of the mobile platform.
The approach uses a simple visual sensor and it has a low memory and
computational footprint. Our approach has the potential to completely augment,
or even supplement, GPS based navigation since it functions ubiquitously in all
environments.
主要な位置特定手法としてGPSがあげられるが，位置精度や適用可能範囲に
課題がある。そこで，GPSが適用できない屋内・屋外環境における新しい位置
特定手法を考案した。この手法には二つの基本要素がある。ひとつは，地理情
報システム
（GIS）
に含まれる屋内・屋外の公的な道路網情報データ，もうひと
つは，車両からの情報に基づく走行軌跡データで，慣性計測センサやビジュア
ルオドメトリを使って推定される。筆者は，道路網情報をモデル化されたグラフ
データ構造で表現した。この構造では，様々なタイプの道路はグラフの線
（エッ
ジ）
，交差点は交点
（ノード）
として表される。このグラフモデル中でリアルタイム
に軌跡を探索し，車両の位置を特定する。この手法は，単純な視覚センサを使
用し，小容量メモリの小型コンピュータ上で動作する。あらゆる環境で場所を選
ばずに適用できることから，本手法は，GPSを使ったナビゲーションを補強し，
さらにデータを補完する技術になると考える。
Introduction
$XWRQRPRXVQDYLJDWLRQLVDQHPHUJLQJWHFKQRORJ\ZLWKD
KXJH SRWHQWLDO VHOIGULYLQJ FDUV DUH DOPRVW URXQG WKH
FRUQHU7KLVWHFKQRORJ\UHTXLUHVDFFXUDWHJHRORFDOL]DWLRQ
LQ UHDOWLPH WR HIIHFWLYHO\ QDYLJDWH XUEDQ HQYLURQPHQWV
&XUUHQWO\QDYLJDWLRQWHFKQRORJLHVDUHWLPHRIÀLJKWUHOL
DQW VXFK DV *OREDO 3RVLWLRQLQJ 6\VWHP *36[1] 8OWUD
:LGH %DQG 8:%[2] %OXHWRRWK[3] DQG DFWLYH UDGLRIUH
TXHQF\ LGHQWL¿FDWLRQ $5),'[4] ZKLFK KDYH PXOWLSOH
LVVXHV7KHDFFXUDF\RIVWDQGDUGWLPHRIÀLJKWGHYLFHVLV
XQDFFHSWDEO\ SRRU IRU WKH SXUSRVHV RI DXWRQRPRXV QDYL
JDWLRQDQGDFFXUDWHVHQVRUVDUHYHU\H[SHQVLYHWRGHSOR\
DQG PDLQWDLQ (VSHFLDOO\ FRPPRQO\ DGRSWHG *36 EDVHG
VROXWLRQ LV XQUHOLDEOH LQ VHYHUDO HQYLURQPHQWV ZKHUH
WUDQVPLVVLRQEHWZHHQWKHGHYLFHDQGVDWHOOLWHVLVLPSHGHG
E\VXUURXQGLQJVWUXFWXUHVOLNHWDOOEXLOGLQJVLQDQXUEDQ
HQYLURQPHQW WKH VR FDOOHG µXUEDQ FDQ\RQV¶ ,Q DGGLWLRQ
*36LVXQDYDLODEOHLQGRRUVXQGHUJURXQGLQWXQQHOVHWF
DQGFDQDOVREHGHJUDGHGRUGHQLHGLQFHUWDLQJHRJUDSKLF
UHJLRQV,QRUGHUWREHYLDEOHIRUDKXJHFRQVXPHUPDUNHW
WKHUHLVDQHHGIRUJHRORFDOL]DWLRQVROXWLRQWKDWRSHUDWHV
ZLWKORZFRVWVHQVRUVDQGSXEOLFGRPDLQGDWD
:H SRVH WKH SUREOHP RI DFFXUDWH JHRORFDOL]DWLRQ DV D
FRPELQDWLRQRIDFFXUDWHUHODWLYHORFDOL]DWLRQZLWKLQDYHU\
Special issue October 2016
37
) HDWXUH$UWLFOH
堀場雅夫賞特別賞受賞者論文
Ubiquitous Geo-Localization Without Infrastructure
Figure 1 Trajectory Geo-Localization: VO is used to compute a trajectory in real-time which is abstracted as a sub-graph and is progressively localized by
sub-graph matching in a graph abstraction of transport network map of the region of interest.
ODUJH VSDWLDO VHDUFK VSDFH ZLWKRXW UHTXLULQJ DQ\ LQIUD
VWUXFWXUHVXFKDVGHSOR\PHQWRIDQWHQQDV7KHUHODWLYHOR
FDOL]DWLRQFDQEHFRPSXWHGXVLQJ9LVXDO2GRPHWU\92
RU ,08 DQG WKH ODUJH VSDWLDO VHDUFK VSDFH LV DFTXLUHG
IURP*,6GDWDEDVHV[5]7KHNH\LQVLJKWLVWKDWWKHPRWLRQ
RIWKHW\SLFDOPRELOHSODWIRUPLVFRUUHODWHGWRWKHDVVRFL
DWHGWRSRORJ\RIWKHVSDWLDOVHDUFKVSDFH)RUH[DPSOHD
FDUPRYHVRQURDGVDQGVRWKHWUDMHFWRU\RIWKHFDUIRUD
¿QLWH GLVWDQFH RI WUDYHO LV FRUUHODWHG WR D VXEVHW RI WKH
WUDQVSRUWQHWZRUNOD\HURI*,6LQWKHVSDWLDOVHDUFKVSDFH
IRUWKDWFDU7KHXVHRIRQHRUPRUHOD\HUVRI*,6EDVHG
RQ WKH W\SH RI PRELOH SODWIRUP DQG VHQVRUV IDFLOLWDWHV
ÀH[LELOLW\LQRXUDSSURDFKDQGWKHDELOLW\WRVFDOHWKHVROX
WLRQ
UHTXLUHVDQLQLWLDOK\SRWKHVLVRIUHJLRQZKHUHWKHPRELOH
SODWIRUP FDQ EH ORFDWHG ZKLFK FDQ EH LQGRRUV RU RXW
GRRUV)RUW\SLFDOXVHFDVHVWKHUHJLRQRILQWHUHVWZRXOG
EHDFRXQWU\RUFLW\VL]HGDUHD7KHDSSURDFKLVLOOXVWUDWHG
LQWKHEORFNGLDJUDPLQFigure 17KH¿UVWVWHSLVWRDF
TXLUH PDS GDWD IURP D SXEOLF GRPDLQ VRXUFH VXFK DV
2SHQ6WUHHW0DS2607KLVGDWDLVW\SLFDOO\JLYHQLQDQ
;0/IRUPDWDQGLVVWUXFWXUHGXVLQJWKUHHEDVLFHQWLWLHV
QRGHVZD\VDQGUHODWLRQV7KHQRGHVUHSUHVHQWSRLQWHOH
PHQWV LQ WKH PDS DQG DUH GH¿QHG E\ WKHLU *36 FRRUGL
QDWHV 7KH LQWHUHVWLQJ PDS VWUXFWXUHV IRU XV DUH VWUHHWV
ZKLFK DUH PRGHOOHG DV ZD\V DQG FDQ EH H[WUDFWHG IURP
WKH ;0/ GDWD LQ WR IRUP D VWUHHW JUDSK RI WKH UHJLRQ RI
LQWHUHVW
Encoding Motion
6LPLODUWRRXWGRRULQGRRUGDWDLVSURYLGHGE\PHDQVRID
EXLOGLQJ LQIRUPDWLRQ PRGHO %,0 DQG FRQWDLQV VLPLODU
OD\HUVRIGDWDZKLFKFDQEHXVHGWRHVWLPDWHWKHORFDWLRQ
RIWKHSODWIRUPLQGRRUV
:HSRVHWKHSUREOHPRIDFFXUDWHPRELOHSODWIRUPJHROR
FDOL]DWLRQDVDFRPELQDWLRQRIDSSUR[LPDWHJHRORFDOL]D
WLRQ DQG DFFXUDWH UHODWLYH ORFDOL]DWLRQ 7KH DSSURDFK
（a）
（b）
Figure 2 (a) Encoding trajectory of mobile platform and edges on graph of transport network. The motion from a point at a sampling interval to next point is
measured in angle which is quantized into bins. (b) Feature descriptor for estimated 3D trajectory of mobile platform and edges of graph representing roads.
38
Special issue October 2016
Technical Reports
7KHVHFRQGVWHSLVWRFRPSXWHIHDWXUHGHVFULSWRUVIRUWKH
' WUDMHFWRU\ RI WKH SODWIRUP DQG WKH WUDQVSRUW QHWZRUN
JUDSK:HFKRVHDFRQWRXUWDQJHQWVKRZQLQFigure 2aDV
WKHXQLIRUPVDPSOHGLVWDQFH,QRUGHUWRDLGVFDODELOLW\RI
WKHVROXWLRQZHRSWHGIRUDGHVFULSWRUWKDWLVERWKIDVWDQG
JRRGIRUVHDUFKWUDMHFWRU\LQWKHJUDSK$WUDMHFWRU\LVDV
VXPHGWREHFRQVLVWHGRISRLQWVDWURXJKO\XQLIRUPVDP
SOLQJGLVWDQFH7KHTXDQWL]DWLRQDVVRFLDWHGZLWKVDPSOLQJ
GLVWDQFHLVUHOHYDQWLQWHUPVRIVHQVLWLYLW\WRWKHPRWLRQRI
WKH PRELOH SODWIRUP 7KH FKRLFH RI TXDQWL]DWLRQ OHYHO LV
EDVHGRQRSWLPL]LQJVHQVLWLYLW\WRPRELOHSODWIRUPPRWLRQ
ZKLOHPLQLPL]LQJHIIHFWRIQRLVHFigure 2bVKRZVDWUD
MHFWRU\HQFRGHGLQWKLVPDQQHU
Trajectory Search in Graph
:LWK WKH DLP RI IDFLOLWDWLQJ D JHQHUDOL]HG VROXWLRQ WKDW
GRHVQRWUHTXLUHNQRZOHGJHRIJHRJUDSKLFFRRUGLQDWHVRU
JOREDO GLUHFWLRQ LQIRUPDWLRQ RI WKH PRELOH SODWIRUP ZH
DVVXPHWKDWWKHWUDMHFWRU\HVWLPDWHGE\92LVLQLWVVLP
SOLVWLF IRUP D VHTXHQFH RI SRLQWV ZLWK UHODWLYH GLVWDQFH
PHDVXUH7KLVPHDQVWKHPRVWUHOLDEOHPRGHRIHQFRGLQJ
WKHWUDMHFWRU\LVFKDQJHLQDQJOHRIPRWLRQRIWKHPRELOH
SODWIRUPZKLFKLVHTXLYDOHQWWRFRQWRXUDQJOHVRIDVSOLQH
DSSUR[LPDWLRQ RI WKH DFTXLUHG VHTXHQFH RI SRLQWV $
JUDSK DEVWUDFWLRQ RI WKLV WUDMHFWRU\ LV FUHDWHG XVLQJ WKH
FRQWUROSRLQWVDVQRGHVRIWKHJUDSKDQGVHWRISRLQWVEH
WZHHQFRQWUROSRLQWVFRPSULVHDQHGJHRIWKHJUDSK6LQFH
WKHPRELOHSODWIRUPWUDYHOVRQWKHURDGQHWZRUNWKHWUDMHF
WRU\JUDSKFDQEHDVVXPHGWREHDVXEJUDSKRIWKHPDS
JUDSK2XUDSSURDFKWRVHDUFKLQJIRUDPDWFKLQWKLVODUJH
JUDSK LV WR LQWURGXFH SDWK JUDSK 6HDUFKLQJ IRU D WUDMHF
WRU\ LV HTXLYDOHQW WR VHDUFK IRU D VXEJUDSK RI WKH ODUJHU
JUDSK6RZHFRPSXWHDVLPLODULW\VFRUHEHWZHHQWKHWUD
MHFWRU\ JUDSK DQG SDWK JUDSK IRU DOO SDWK JUDSKV LQ WKH
PDSJUDSK7KHEHVWPDWFKLQJSDWKJUDSKLVWKHQVHOHFWHG
DVWKHVROXWLRQ
Experiments
:HIRXQGWKDWDEVWUDFWLRQRIWUDMHFWRU\DQGPDSWRJUDSK
VWUXFWXUH VKRXOG EH EDVHG RQ WKH VFDOH DQG IUHTXHQF\ RI
RFFXUUHQFH RI VLJQL¿FDQW IHDWXUHV 7R LPSURYH HI¿FLHQF\
RIWKHLPSOHPHQWDWLRQZHEHJLQZLWKDQDQDO\VLVRIWKHWR
SRORJ\ RI WKH PDS GDWD 7KHUH UHVXOWV ZHUH XVHG WR WXQH
WKH SDUDPHWHUV RI RXU JHRORFDOL]DWLRQ SLSHOLQH ZKLFK LV
HYDOXDWHG RQ VHYHUDO GDWDVHWV IRU GLIIHUHQW SDUWV RI WKH
ZRUOG
:H DFTXLUHG PDSV IRU UHJLRQV RI LQWHUHVW IURP
2SHQ6WUHHW0DS WR HYDOXDWH WKH SHUIRUPDQFH RI RXU
DSSURDFKIRUGLIIHUHQWW\SHVRIWUDQVSRUWQHWZRUNWRSROR
JLHVLQFOXGLQJXUEDQVHPLXUEDQDQGFRXQWU\URDGVDQG
VFDOHRIWKHVHDUFKUHJLRQ:HVHOHFWHGPDSVIRUµ)UDQNOLQ
FRXQW\2KLR¶ZKLFKSURYLGHVDW\SLFDO86FRXQW\VL]HG
PDS µ:DVKLQJWRQ '&¶ ZKLFK LV D KLJKO\ XUEDQ UHJLRQ
ZLWK D ODWWLFH JULG VWUXFWXUHG URDG QHWZRUN DQG
µ0RQWSHOOLHU)UDQFH¶ZKLFKLVDVHPLXUEDQDQGFRXQWU\
VSDJKHWWL VKDSHG URDG QHWZRUN RI D VWDWH VL]HG UHJLRQ
6RPHRIWKHUHVXOWVRIRXUH[SHULPHQWVIRUHDFKRIWKHVH
UHJLRQVLVLOOXVWUDWHGLQFigure 3,QLWLDOO\WKHHVWLPDWHG
WUDMHFWRU\LVVKRUWDVWKHPRELOHSODWIRUPEHJLQVPRYLQJ
:HFRPSXWHDVLPLODULW\VFRUHDQGREVHUYHWKDWWKHUHDUH
VHYHUDOPDWFKLQJSDWKVVKRZQLQWKH¿UVWFROXPQRIWKH
)LJXUHIRUHDFKUHJLRQ$VWKHHVWLPDWHGWUDMHFWRU\JURZV
WKHQXPEHURIFDQGLGDWHSDWKVUHGXFHEDVHGRQRXUHPSLU
LFDOO\ GHWHUPLQHG WKUHVKROG VKRZQ LQ WKH FHQWHU FROXPQ
LQWKH¿JXUH6XEVHTXHQWO\DVLQJOHSDWKZLWKWKHKLJKHVW
VLPLODULW\VFRUHUHPDLQVZKLFKLVRXUPDWFKHGSDWKLQWKH
PDSJUDSK7KHULJKWFROXPQIRUHDFKUHJLRQLQWKH¿JXUH
VKRZVWKHPDWFKHGSDWKLQWKHJUDSKIRUWKHFRUUHVSRQG
LQJ PRELOH SODWIRUP WUDMHFWRU\ ,Q RXU H[SHULPHQWV ZH
IRXQGWKDWTXLFNVHDUFKGHSHQGVRQWKHXQLTXHQHVVRIWKH
WUDMHFWRU\DQGWKHQDWXUHRIWKHPDSJUDSK
Conclusions
:HFRQVLGHUWKHSUREOHPRIJHRORFDOL]DWLRQLQUHDOWLPH
1RWH WKDW WKH HVWLPDWHG WUDMHFWRU\ FRQWLQXRXVO\ JURZV
ZLWK WLPH DQG FRQVHTXHQWO\ WKH WUDMHFWRU\ JUDSK DOVR
JURZV ZLWK WLPH &RQVHTXHQWO\ ZH IRUPXODWH RXU WUDMHF
WRU\ VHDUFK VROXWLRQ DFFRUGLQJ WR D VXEVWULQJ PDWFKLQJ
DOJRULWKP[6] 7KH WUDMHFWRU\ LV D VHTXHQFH RI TXDQWL]HG
HGJHRULHQWDWLRQVZHGHVFULEHGHDUOLHU(DFKJUDSKSDWKLV
DOVR VLPLODUO\ HQFRGHG DV D VHTXHQFH RI TXDQWL]HG HGJH
RULHQWDWLRQV 7KH NH\ EHQH¿W RI XVLQJ VWULQJ PDWFKLQJ LV
WKDWSDUWLDOPDWFKHVDOVRSURGXFHDUHDVRQDEO\JRRGVLPL
ODULW\ VFRUH 7KLV IRUPXODWLRQ PDNHV UREXVW WR QRLVH
PLVVLQJ GDWD HUURQHRXV YHKLFOH PRWLRQ OLNH VZHUYLQJ
PDS UHJLVWUDWLRQ ÀDZV HWF 7KH PRWLYDWLRQ IRU WKLV DS
SURDFKLVGHULYHGIURPWKHSRSXODU%DJRI:RUGVPHWKRG
LQ FRPSXWHU YLVLRQ WKDW KDV DOVR EHHQ XVHG IRU
6LPXOWDQHRXV/RFDOL]DWLRQDQG0DSSLQJ
$QRYHODSSURDFKIRUJHRORFDOL]LQJSRVLWLRQRIDPRELOH
SODWIRUP LV SUHVHQWHG :H KDYH SURSRVHG D PRGL¿HG DS
SURDFK WR FODVVLFDO YLVXDO RGRPHWU\ SLSHOLQH ZLWK LQGRRU
DQG RXWGRRU PDS GDWD LQ D XQL¿HG IUDPHZRUN :H KDYH
GHVFULEHG WKH DOJRULWKPLF DQG LPSOHPHQWDWLRQ GHWDLOV RI
RXUPHWKRGDQGGHPRQVWUDWHGLWRQVHYHUDOGLIIHUHQWW\SHV
RI PDSV IURP GLIIHUHQW UHJLRQV RI WKH ZRUOG :H KDYH
GHPRQVWUDWHG H[FHOOHQW UHVXOWV RQ RXU SURSRVHG YLVXDO
RGRPHWU\ SLSHOLQH RQ D EHQFKPDUN GDWDVHW 2XU UHVXOWV
VKRZWKDWWKHSURSRVHGV\VWHPLVDEOHWRSURYLGHIXOO\DX
WRPDWLF JOREDO ORFDOL]DWLRQ DW D ORZ LQIUDVWUXFWXUDO FRVW
%XLOGLQJRQWKLVZRUNZHSODQWRIXUWKHULQYHVWLJDWHWKH
XVHRIPDSVHVSHFLDOO\GLIIHUHQWOD\HUVRID*,6GDWDEDVH
VXFKDVEXLOGLQJK\GURORJ\UHOLHIPDSVHWF:HDOVRSODQ
WR LQWHJUDWH RXWGRRU DQG LQGRRU *36GHQLHG QDYLJDWLRQ
Special issue October 2016
39
) HDWXUH$UWLFOH
堀場雅夫賞特別賞受賞者論文
Ubiquitous Geo-Localization Without Infrastructure
（a）
（b）
（c）
Figure 3 Geo-localization of mobile platform trajectory. (from top to bottom) For the geographic regions of ‘Franklin County, Ohio’, ‘Washington DC, US’,
and ‘Montpellier, France’. The estimated trajectory of the mobile platform is shown on the top left; the histogram on the top right shows the match
score of each candidate path; and the map shows the current candidate paths. In the first column, the initial trajectory is small and several candidate paths are spawned. The similarity match score is shown in the histogram graph. As trajectory grows several candidates are pruned. Finally,
the best matching path remains, shown in the right column, which is used for inferring geo-location of the mobile platform.
40
Special issue October 2016
Technical Reports
IRUVHDPOHVVXQLQWHUUXSWHGJHRQDYLJDWLRQLQDOOORFDWLRQV
DQGHQYLURQPHQWFRQGLWLRQV
Acknowledgements
, ZRXOG OLNH WR H[SUHVV P\ JUDWLWXGH IRU WKH PHPEHUV RI
P\ UHVHDUFK /DE SDUWLFXODUO\ $VKLVK *XSWD 6LDYDVK
+RVVHLQ\$ODPGDU\DQG<XMLD=KDQJDW268ZKRKHOSHG
GHYHORSPHQW DQG UHVHDUFK RI WKLV V\VWHP 7KLV V\VWHP
FRXOGQRWKDYHEHHQGHYHORSHGZLWKRXWWKHVXSSRUWRI86
'HSDUWPHQWRI'HIHQVH'HIHQVH,QWHOOLJHQFH$JHQF\DQG
1DWLRQDO$HURQDXWLFVDQG6SDFH$JHQF\
References
[ 1 ] 1DWLRQDO 5HVHDUFK &RXQFLO 86 &RPPLWWHH 5HSRUW RQ WKH )XWXUH
RI WKH *OREDO 3RVLWLRQLQJ 6\VWHP 1DWLRQDO $FDGHP\ RI 3XEOLF
$GPLQLVWUDWLRQ 7KH JOREDO SRVLWLRQLQJ V\VWHP D VKDUHG QD
WLRQDO DVVHW UHFRPPHQGDWLRQV IRU WHFKQLFDO LPSURYHPHQWV DQG HQ
KDQFHPHQWV1DWLRQDO$FDGHPLHV3UHVVS,6%1
[ 2 ] = KRX</DZ&/*XDQ</&KLQ),QGRRUHOOLSWLFDOORFDOL]D
WLRQ EDVHG RQ DV\QFKURQRXV 8:% UDQJH PHDVXUHPHQW ,(((
7UDQVDFWLRQVRQ,QVWUXPHQWDWLRQDQG0HDVXUHPHQW
[ 3 ] -LP(GZDUGV$SSOH,V/DXQFKLQJ$9DVW3URMHFW7R0DS7KH,QVLGH
2I(YHU\/DUJH%XLOGLQJ,W&DQ%XVLQHVV,QVLGHU-XQH
[ 4 ] 5H]D $: *HRN 7. ,QYHVWLJDWLRQ RI LQGRRU ORFDWLRQ VHQVLQJ YLD
5),' UHDGHU QHWZRUN XWLOL]LQJ JULG FRYHULQJ DOJRULWKP :LUHOHVV
3HUVRQDO&RPPXQLFDWLRQV
[ 5 ] 2 SHQ6WUHHW0DS)RXQGDWLRQ:LNLFRQWULEXWRUV0DLQ3DJH>,QWHUQHW@
2SHQ6WUHHW0DS )RXQGDWLRQ :LNL $XJ 87& >FLWHG
$XJ @ $YDLODEOH IURP KWWSZLNLRVPIRXQGDWLRQRUJZ
LQGH[SKS"WLWOH 0DLQB3DJHROGLG [ 6 ] 3DXO5DQG1HZPDQ3)$%0$3'7RSRORJLFDOPDSSLQJ
ZLWK VSDWLDO DQG YLVXDO DSSHDUDQFH ,Q ,((( ,QWHUQDWLRQDO
&RQIHUHQFHRQ5RERWLFVDQG$XWRPDWLRQSS±
Alper YILMAZ
'HSDUWPHQWRI&LYLO
(QYLURQPHQWDODQG*HRGHWLF(QJLQHHULQJ
7KH2KLR6WDWH8QLYHUVLW\
$VVRFLDWH3URIHVVRU
3K'
Special issue October 2016
41

Download Report