...

様々な歩行状況下における歩容認証手法の性能評価

by user

on
Category: Documents
22

views

Report

Comments

Transcript

様々な歩行状況下における歩容認証手法の性能評価
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
༷ʑͳาߦঢ়‫گ‬Լʹ͓͚Δา༰ೝূख๏ͷੑೳධՁ
౦ࢁ ါਅ1,a)
ᴳ‫ ݪ‬༃1,b)
੢໺ ߃2,c)
ീ໦ ߁࢙1,d)
֓ཁɿۙ೥ɼࢦ໲ೝূɼ೒࠼ೝূͳͲʹଓ͘৽͍͠όΠΦϝτϦοΫೝূͱͯ͠ɼηϯα͔Β཭Εͨ৔ॴ
Ͱ΋ຊਓ֬ೝ͕Մೳͳา༰ೝূʹߴ͍ؔ৺͕ू·͍ͬͯΔɽ͜Ε·Ͱʹଟ͘ͷา༰ೝূख๏͕ఏҊ͞Εͯ
͍Δ͕ɼͦΕΒͷख๏ʹର͢Δ༷ʑͳঢ়‫گ‬Λ૝ఆͨ͠แ‫ׅ‬తͳੑೳධՁ͸ະͩͳ͞Ε͍ͯͳ͍ɽͦ͜Ͱɼ
ຊ‫Ͱڀݚ‬͸ɼ༷ʑͳঢ়‫گ‬Լʹ͓͍ͯɼา༰ೝূख๏ͷੑೳධՁΛߦ͏͜ͱΛ໨తͱ͢ΔɽੑೳධՁʹ༻͍
Δา༰ಛ௃͸ɼγϧΤοτʹ‫ͮ͘ج‬࿡ͭͷา༰ಛ௃ʹՃ͑ͯɼಈ͖৘ใΛΑΓੵ‫ۃ‬తʹར༻ͨ͠า༰ಛ௃
Ͱ͋ΔϑϨʔϜؒࠩ෼ʹ‫ཱͮ͘ج‬ମߴ࣍‫ݾࣗॴہ‬૬ؔಛ௃΍ɼΦϓςΟΧϧϑϩʔʹ‫ ͮ͘ج‬Gait motion
descriptors Λར༻͢Δɽ·ͨɼಛ௃ͷর߹ख๏ͱͯ͠͸ɼϢʔΫϦου‫ͼٴ཭ڑ‬ਖ਼४൑ผ෼ੳΛ༻͍Δɽ
࣮‫Ͱݧ‬͸ɼ଎౓͕มԽ͢Δ৔߹ɼ෰૷͕มԽ͢Δ৔߹ͷೋͭͷঢ়‫͚͓ʹگ‬Δา༰ೝূख๏ͷੑೳධՁΛߦ
͏ɽ࠷‫ʹޙ‬ɼಘΒΕͨ݁ՌΛߟ࡯͠ɼ֤า༰ೝূख๏ͷ༗ޮੑΛൺֱͨ͠ɽ
Ωʔϫʔυɿา༰ಛ௃ɼੑೳධՁɼ଎౓มԽɼ෰૷มԽɼর߹ख๏
1. ͸͡Ίʹ
ۙ೥ɼ‫ݸ‬ਓ৘ใอ‫ޢ‬๏ͷࢪߦ΍஌తࡒ࢈ͷ૿ՃʹΑΓɼ
ϝτϦοΫೝূΛར༻͢ΔϓϩδΣΫτ [1] ͕ਐߦ͍ͯ͠
Δɽ͜Ε͸ɼόΠΦϝτϦοΫೝূͷར఺Ͱ͋Δɼ‫ޡ‬ೝࣝ
཰ͷ௿͞΍ɼඃೝূऀʹ͔͔Δෛ୲͕͕ܰ͞ɼΠϯυͷΑ
‫ݸ‬ਓೝূγεςϜʹ͓͚ΔηΩϡϦςΟͷ‫ڧ‬Խ͕ॏཁࢹ͞
͏ͳ৽‫ͯͬͱʹࠃڵ‬ड༰͠΍͍͢͜ͱ͕ཧ༝ͱͯ͠‫͛ڍ‬Β
Ε͍ͯΔɽ࠷ۙͰ͸ɼԕִૢ࡞΢ΟϧεʹΑΔ୺຤ͷ৐ͬ
ΕΔɽଞʹɼࢦ໲ʹΑΔ‫ػࢉܭ‬΍ϞόΠϧ୺຤΁ͷϩάΠ
औΓ΍ɼѱ࣭ͳϋοΩϯάʹΑΔ‫ݸ‬ਓ৘ใͷྲྀग़ͱ͍ͬͨ
ϯγεςϜɼ੩຺ʹΑΔΞΫηείϯτϩʔϧγεςϜɼ
αΠόʔࣄ͕݅ଟൃ͓ͯ͠ΓɼηΩϡϦςΟ‫ڴ‬Җ͸΋͸΍
ࢦ໲΍ DNA ʹΑΔ൜ࡑ૞ࠪͷͨΊͷؑఆγεςϜ౳Ͱར
ଞਓࣄͰ͸ͳ͍ͱ͜Ζ·Ͱഭ͖͍ͬͯͯΔɽͦͷΑ͏ͳഎ
༻͞Ε͍ͯΔɽ·ͨɼ೔ຊࠃ಺ʹ͓͍ͯ͸ɼۜߦ ATM ʹ
‫ܠ‬ͷதɼਓͷੜମ৘ใΛར༻ͨ͠όΠΦϝτϦοΫೝূ͕
͓͍ͯɼ҉ূ൪߸ʹՃ͑ͯࢦͷ੩຺ύλʔϯͷ৘ใΛར༻
ߴ͍ؔ৺ΛूΊ͍ͯΔɽ
͢Δ͜ͱͰηΩϡϦςΟͷ‫ڧ‬ԽΛਤ͍ͬͯΔɽ
όΠΦϝτϦοΫೝূͱ͸ɼࢦ໲΍‫ٿ؟‬ͷ೒࠼ɼ੩຺ͱ
͔͠͠ͳ͕Βɼࢦ໲ɼ‫ٿ؟‬ͷ೒࠼ɼ੠໲ͱ͍ͬͨ਎ମత
͍ͬͨ਎ମతಛ௃΍ॺ໊ͱ͍ͬͨߦಈతಛ௃ʹΑͬͯຊਓ
ಛ௃Λར༻ͨ͠όΠτϝτϦΫεೝূ͸ɼηϯα͔Βͷ‫ڑ‬
֬ೝΛߦ͏ೝূํࣜͷ͜ͱͰɼੜମ৘ใ͕΋ͭීวੑɼ།
཭͕ۙ͘ͳ͚Ε͹ͳΒͳ͍఺΍ඃೝূऀ͕ࣗΒొ࿥ʹ޲͔
ҰੑɼӬଓੑͱ͍ͬͨಛ௃Λར༻͢Δ͜ͱ͔Βɼ҉ূ൪߸
Θͳ͚Ε͹ͳΒͳ͍ͱ͍͏ܽ఺͕͋Δɽ͜ͷΑ͏ͳܽ఺Λ
΍ύεϫʔυʹൺ΂‫ݪ‬ཧతʹ‫ͳݻڧ‬ೝূٕज़Ͱ͋Δͱߟ͑
ิ͏όΠΦϝτϦοΫೝূͱͯ͠ɼηϯα͔Β཭ΕͨҐஔ
ΒΕ͍ͯΔɽΞϝϦΧ߹ऺࠃͷύεϙʔτίϯτϩʔϧʹ
ͰೝূΛߦ͏͜ͱ͕Ͱ͖Δɼਓͷา͖ํͷ‫ݸ‬ੑʹ‫ͮ͘ج‬า
͓͍ͯ͸ɼࢦ໲‫إͼٴ‬ը૾Λऔಘ͢Δ͜ͱͰɼςϩϦετ
༰ೝূ͕஫໨ΛूΊ͓ͯΓɼ๷൜ΧϝϥΛར༻ͨ͠޿Ҭ‫؂‬
΍ࢦ໊ख഑൜ͷೖࠃΛ๷͙ࢼΈ͕ͳ͞Ε͍ͯΔɽΠϯυʹ
ࢹ΍൜ࡑ૞ࠪ΁ͷԠ༻͕‫ظ‬଴͞Ε͍ͯΔɽ࣮ࡍʹɼΠΪϦ
͓͍ͯ͸ɼ12 ԯʹ΋্Δ๲େͳ਺ͷࠃຽͷ‫ݸ‬ਓΛೝূ͢
εͰ͸ɼ‫ڧ‬౪൜ʹର͢Δา༰ೝূͷ݁Ռ͕ࡋ൑ॴʹ͓͚Δ
Δख๏ͱͯ͠ɼࢦ໲ೝূɼ೒࠼ೝূɼ‫إ‬ೝূͳͲͷόΠΦ
ূ‫͞༻࠾ͯ͠ͱڌ‬Εͨࣄྫ [2] ͕͋Γɼ·ͨɼ೔ຊʹ͓͍
ͯ΋ɼา༰ೝূʹΑΔؑఆ݁Ռ͕൜ࡑ૞ࠪࢧԉʹ‫͞༻׆‬Ε
1
2
a)
b)
c)
d)
େࡕେֶ
Osaka University
υϨΫηϧେֶ
Drexel University
[email protected]
[email protected]
[email protected]
[email protected]
ⓒ 2013 Information Processing Society of Japan
ͨࣄྫ΋͋Δɽ
͜ͷΑ͏ͳഎ‫ܠ‬ͷԼɼ͜Ε·Ͱଟ਺ͷา༰ೝূख๏͕ఏ
Ҋ͞Ε͖͓ͯͯΓɼͦΕΒ͸େ͖͘ϞσϧϕʔεͱΞϐΞ
ϥϯεϕʔεͷ 2 ख๏ʹ෼͚Δ͜ͱ͕Ͱ͖Δɽ
Ϟσϧϕʔεͷख๏Ͱ͸ɼೖྗը૾ʹϞσϧΛ౰ͯ͸ΊΔ
1
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
͜ͱͰɼਓͷମ‫ܕ‬΍ಈ͖ͱ͍ͬͨಛ௃Λநग़͢ΔɽUrtasun
Β [3] ͸ɼղ๤ֶతͳਓମϞσϧΛ౰ͯ͸Ίɼؔઅ֯౓ͷ
நग़Λߦ͍ɼSpencer Β [4] ͸ɼநग़͞Εͨؔઅ఺ͷप‫ظ‬త
ͳҐஔ৘ใ͔Βɼ‫؍‬ଌํ޲ʹରͯ͠ෆมͳؔઅ֯౓ͷநग़
Λߦͬͨɽ͔͠͠ͳ͕ΒɼϞσϧϕʔεͷख๏͸ɼ௿ղ૾
౓ͷը૾ʹରͯ͠ɼϞσϧͷ౰ͯ͸Ί΍ಛ௃நग़ʹࣦഊ͢
(a) GEI
(b) FDF
(c) GENI
Δ৔߹͕͋Δɽ
ΞϐΞϥϯεϕʔεͷख๏Ͱ͸ɼը૾Λ௚઀ղੳ͠ಛ
௃Λநग़͢ΔɽHan ͱ Bhanu[5] ͸γϧΤοτը૾ྻΛप
‫Ͱظ‬ฏ‫ۉ‬Խͨ͠ɼGait energy image (GEI) ΛఏҊͨ͠ɽ
Makihara Β [6] ͸γϧΤοτը૾Λ্࣌ؒ࣠ʹੵΈॏͶΔ
(d) MGEI
(e) GFI
(f) CGI
͜ͱͰಘΒΕΔา༰γϧΤοτϘϦϡʔϜ͔Βɼ࣌ؒඇґ
ଘͷप೾਺ྖҬಛ௃ (Frequency domain feature, FDF) Λ
நग़ͨ͠ɽBashir Β [8] ͸ 1 प‫ظ‬෼ͷγϧΤοτը૾ྻ͔
Β࣌ؒతͳΤϯτϩϐʔΛ‫ͨ͠ࢉܭ‬ɼGait entropy image
(GEnI) ΛఏҊͨ͠ɽLam Β [9] ͸γϧΤοτը૾ؒͷΦ
ϓςΟΧϧϑϩʔΛ‫ͨ͠ࢉܭ‬ɼGait flow image (GFI) Λ
(g) CHLAC
ఏҊͨ͠ɽBashir Β [10] ͸ GEnI ͔Β࡞੒ͨ͠ϚεΫʹ
GEI Λ౰ͯΔɼMasked GEI based on GEnI (MGEI) Λఏ
Ҋͨ͠ɽWang Β [11] ͸γϧΤοτͷྠֲΛҐ૬ʹ‫͍ͮج‬
ͯ৭͚ͮͯ͠ 4 ෼ͷ 1 प‫ظ‬ຖʹॏͶ߹ΘͤΔ Chrono-gait
image (CGI) ΛఏҊͨ͠ɽ͔͠͠ͳ͕ΒɼγϧΤοτʹ‫ج‬
(h) GMD
า༰ಛ௃
ͮ͘ಛ௃நग़͸ɼ੩తͳ෦෼ͱಈతͳ෦෼Λࠞࡶͯ͠ѻͬ
ਤ 1
͍ͯΔͨΊɼମ‫ʹܕ‬େ͖͘ґଘ͢Δ܏޲͕͋Δɽ
Ұ ํ ɼγ ϧ Τ ο τ ը ૾ Λ ༻ ͍ ͳ ͍ า ༰ ಛ ௃ ͱ ͠ ͯ ɼ
Kobayashi ͱ Otsu[12] ͸ɼϑϨʔϜؒࠩ෼ʹΑΓಈతͳಛ
௃఺Λऔಘ͠ɼ251 ‫ݸ‬ͷۭ࣌ؒύλʔϯΛநग़͢Δɼཱମߴ
ख๏ͱͯ͠ɼϢʔΫϦου‫ͮ͘جʹ཭ڑ‬র߹ɼਖ਼४൑ผ෼
ੳʹ‫ͮ͘ج‬র߹ͷ 2 ख๏Λ༻͍Δɽ
࣍‫ݾࣗॴہ‬૬ؔ (Cubic higher-order local auto-corelation,
2. า༰ಛ௃
CHLAC) ΛఏҊͨ͠ɽ·ͨɼBashir Β [13] ͸ɼೖྗը૾ؒ
2.1 γϧΤοτʹ‫ͮ͘ج‬า༰ಛ௃
ͷΦϓςΟΧϧϑϩʔΛ‫͠ࢉܭ‬ɼಛ௃఺Λ্Լࠨӈͷಈ఺ɾ
ΞϐΞϥϯεϕʔεͷา༰ಛ௃ͷதͰ΋ɼา༰γϧΤο
੩ࢭ఺ͷ‫ͭޒ‬ͷ‫ه‬ड़ࢠʹ෼ผ͢ΔɼGait motion descriptor
τը૾͔Βಛ௃நग़Λߦ͏ํ๏͕‫ࡏݱ‬ͷओྲྀͱͳ͍ͬͯ
(GMD) ΛఏҊͨ͠ɽ
Δɽา༰γϧΤοτը૾͸ɼഎ‫ࠩܠ‬෼ʹ‫ͮ͘ج‬άϥϑΧο
͜ͷΑ͏ʹ༷ʑͳา༰ೝূख๏͕ఏҊ͞ΕΔதɼIwama
Β [14] ͸ΞϐΞϥϯεϕʔεͷา༰ೝূख๏ʹ͍ͭͯɼ
τྖҬ෼ׂ [7] ͳͲΛ༻͍ͯ࡞੒͞ΕΔɽ
·ͨɼา༰ೝূ͸ಈը૾Λର৅ͱ͢ΔͷͰɼর߹ͷࡍʹ
4,000 ਓҎ্ͷେ‫ن‬໛σʔλϕʔεΛ༻͍ͨੑೳධՁΛ
͸ಉҰ࢟੎ಉ࢜ͷ੩ࢭըͰରԠͤ͞ΔͨΊͷϑϨʔϜಉ‫ظ‬
ߦͬͨɽ͔͠͠ͳ͕Βɼ্‫ه‬ͷੑೳධՁ͸ࡾͭͷ‫ݶͰ఺؍‬
ॲཧ͕ඞཁͰ͋Δɽͦ͜Ͱɼ1 प‫ظ‬෼ͷಈը૾͔Β։࢝ϑ
ఆతͰ͋Δͱ͍͏໰୊఺͕͋ͬͨɽୈҰʹɼಉঢ়‫گ‬ԼͰͷ
ϨʔϜͷาߦ࢟੎ʹґΒͳ͍า༰ಛ௃Λऔಘ͠ɼͦͷप‫ظ‬
ੑೳධՁ͔͠ߦΘΕ͍ͯͳ͍͜ͱɼୈೋʹɼγϧΤοτ
୯Ґͷಛ௃ຖʹর߹Λߦ͏͜ͱͰϑϨʔϜಉ‫ॲظ‬ཧΛ؆୯
ը૾ʹ‫ͮ͘ج‬࿡ͭͷา༰ಛ௃ (GEIɾFDFɾGEnIɾGFIɾ
Խ͢Δɽप‫ݕظ‬ग़ͷख๏ʹ͍ͭͯ͸ɼ[6] ͰఏҊ͞Ε͍ͯΔ
MGEIɾCGI) ʹ͍ͭͯͷΈͷੑೳධՁʹ‫ݶ‬ఆ͞Ε͍ͯΔ
γϧΤοτը૾ྻͷਖ਼‫ن‬Խࣗ‫ݾ‬૬ؔ࠷େԽʹ‫ͮ͘ج‬ख๏Λ
͜ͱɼୈࡾʹɼϢʔΫϦου‫ͮ͘جʹ཭ڑ‬୯Ұͷর߹ख๏
༻͍ΔɽҎ߱ɼຊ‫͍༻Ͱڀݚ‬Δา༰ಛ௃ʹ͍ͭͯɼ؆୯ʹ
ʹ‫ݶ‬ఆ͞Ε͍ͯΔ͜ͱ͕‫͛ڍ‬ΒΕΔɽ
આ໌͓ͯ͘͠ɽ
Αͬͯຊ‫Ͱڀݚ‬͸ɼҎԼͷࡾ఺ʹΑΓɼΑΓแ‫ׅ‬తͳา
GEI ͸ɼγϧΤοτը૾ྻΛาߦप‫Ͱظ‬ฏ‫ۉ‬Խ͢Δ͜ͱ
༰ೝূख๏ͷੑೳධՁΛߦ͏͜ͱΛ໨తͱ͢Δɽ(1) ଎౓
ͰಘΒΕΔɽGEI ͸ਓ෺ྖҬ෦෼ɼͭ·Γମ‫ܕ‬ͷେ͖͕͞
มԽ΍෰૷มԽͱ͍༷ͬͨʑͳঢ়‫گ‬Լʹ͓͍ͯੑೳධՁΛ
ಛ௃ʹେ͖͘࡞༻͢ΔͷͰɼಉঢ়‫گ‬ԼͰͷর߹ʹ͓͍ͯߴ
ߦ͏ɽ(2) γϧΤοτʹ‫ͮ͘ج‬า༰ಛ௃ʹՃ͑ɼϑϨʔϜ
͍ੑೳΛൃ‫͢ش‬Δ [14]ɽ͔͠͠ͳ͕Βɼ෰૷มԽͷΑ͏ͳ
ؒࠩ෼ɾΦϓςΟΧϧϑϩʔͱ͍ͬͨɼ੩ࢭ෦෼ͱಈ͖෦
ਓ෺ྖҬͷγϧΤοτ͕େ͖͘มԽ͢Δ৔߹ʹ͸ਫ਼౓͕
෼Λ۠ผͨ͠า༰ಛ௃ʹ͍ͭͯੑೳධՁΛߦ͏ɽ(3) র߹
௿͘ͳΔ܏޲͕͋Δɽਤ 1(a) ͸ GEI ΛՄࢹԽͨ͠ը૾Ͱ
ⓒ 2013 Information Processing Society of Japan
2
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
͋Δɽ
Ϋηϧʹ͍ͭͯɼۭ࣌ؒతͳߴ࣍‫ॴہ‬ύλʔϯͱͷ૬ؔΛ
FDF ͸ɼาߦप‫ͮ͘جʹظ‬प೾਺ྖҬಛ௃Λา༰ಛ௃
‫͠ࢉܭ‬ɼ֤ύλʔϯʹର͢ΔώετάϥϜΛ࡞੒͠ɼͦΕ
ͱͯ͠ར༻͢Δɽप‫ظ‬ຖʹ෼ׂͨ͠γϧΤοτը૾ྻʹର
Λಛ௃ͱ͢ΔɽCHLAC ͸‫ݾࣗॴہ‬૬ؔಛ௃ͷੑ࣭Ͱ͋Δ
ͯ͠ɼ࣌ؒ࣠ํ޲ͷ 1 ࣍‫ࢄ཭ݩ‬ϑʔϦΤม‫׵‬Λ‫͠ࢉܭ‬ɼา
Ճ๏ੑ͕͋Δ͜ͱ͔Βɼর߹ͷࡍʹಈը૾͔Βਓ෺ྖҬΛ
ߦप‫Ͱظ‬ਖ਼‫ن‬Խ͞Εͨ௿प೾੒෼ͷৼ෯εϖΫτϧΛநग़
நग़͢Δඞཁ͕ͳ͍ɽ·ͨɼଞͷา༰ಛ௃ʹൺ΂ͯ௿࣍‫ݩ‬
͢Δ͜ͱͰಘΒΕΔɽFDF ͸ GEI ಉ༷ɼਓ෺ྖҬ͕ಛ௃
ͷಛ௃औಘ͢Δ͜ͱ͕Ͱ͖Δɽਤ 1(g) ͸ CHLAC ͷώε
ʹେ͖͘࡞༻͢Δɽಛʹ 0 ഒप೾੒෼͸ GEI ͱಉ༷ͷ‫ܭ‬
τάϥϜͰ͋Δɽ
ࢉࣜʹͳΔͨΊɼGEI ͱಉ༷ͷಛ௃͕நग़͞ΕΔɽਤ 1(b)
͸ FDF ΛՄࢹԽͨ͠ը૾Ͱ͋Δɽ
GEnI ͸ɼGEI ʹର͢ΔγϟϊϯΤϯτϩϐʔΛࢉग़͢
2.3 ΦϓςΟΧϧϑϩʔʹ‫ͮ͘ج‬า༰ಛ௃
าߦʹ͓͚Δಈతͳ෦෼Λநग़͢Δख๏ͱͯ͠ɼΦϓ
Δ͜ͱͰಘΒΕΔɽGEnI ͸ମ‫װ‬෦෼ͳͲͷप‫Ͱ಺ظ‬લ‫ܠ‬
ςΟΧϧϑϩʔʹ‫ͮ͘ج‬ख๏͕͋Γɼͦͷ୅දྫͱͯ͠ɼ
Ͱ͋Γଓ͚ΔྖҬ΍ɼഎ‫͋Ͱܠ‬Γଓ͚ΔྖҬɼଈͪɼGEI
GMD ͕͋Δɽ
ʹ͓͍ͯന΍ࠇͱͳΔྖҬʹ͓͚Δ஋͸ 0 ͱͳΓɼख଍ͳ
GFI ͷΑ͏ʹγϧΤοτը૾ʹ‫͢ࢉܭ͍ͯͮج‬ΔΦϓ
Ͳͷಈ͖ͷ͋ΔྖҬɼଈͪɼGEI ʹ͓͍ͯփ৭ͱͳΔྖҬ
ςΟΧϧϑϩʔͱ͸ҟͳΓɼGMD ͸ମ಺෦ͷςΫενϟ
Ͱେ͖ͳ஋ͱͳΔΑ͏ͳಛ௃Ͱ͋ΔɽΑͬͯ GEI ͱൺ΂
΋༻͍Δ͜ͱ͔ΒɼΑΓ๛෋Ͱ࣮ࡍతͳಈ͖ಛ௃ΛಘΔ͜
ͯɼΑΓಈ͖Λॏࢹͨ͠ಛ௃Λநग़͢Δ͜ͱ͕Ͱ͖Δɽͦ
ͱ͕Ͱ͖Δɽ·ͨɼGMD Ͱ͸γϧΤοτը૾ΛϚεΫͱ
ͷੑ࣭͔Βɼ෰૷มԽͷΑ͏ͳਓ෺ྖҬ͕େ͖͘มԽ͢Δ
ͯ͠༻͍Δ͜ͱͰࠨӈ্Լํ޲ͷಈ͖෦෼ͱɼͦΕҎ֎ͷ
ঢ়‫Ͱگ‬΋ൺֱతর߹͕͏·͍͘͘Մೳੑ͕͋Δɽਤ 1(c) ͸
੩ࢭ෦෼Λ෼཭ͯ͠औಘ͢Δ͜ͱ͕ՄೳͰ͋Δɽਤ 1(h) ͸
GEnI ΛՄࢹԽͨ͠ը૾Ͱ͋Δɽ
MGEI ͸্‫ه‬ͷ GEIɼGEnI Λ૊Έ߹Θͤͨา༰ಛ௃Ͱ
GMD ͷ੩ࢭ෦෼ɼࠨɾӈɾ্ɾԼํ޲ͷಈ͖෦෼ΛՄࢹ
Խͨ͠ը૾Ͱ͋Δɽ
͋ΔɽGEnI Ͱେ͖ͳ஋Λ΋ͭྖҬɼଈͪɼಈ͖ͷେ͖ͳ
GMD ͸ɼา༰ಛ௃Λ੩తͳ෦෼ͱಈతͳ෦෼ʹ۠ผ͢
ྖҬΛϚεΫͱͯ͠औΓग़͠ɼͦͷϚεΫΛ GEI ʹద༻
ΔͨΊɼ଎౓มԽ΍෰૷มԽͳͲɼ༷ʑͳঢ়‫گ‬มԽʹର͠
͢Δ͜ͱͰಛ௃Λநग़͢ΔɽMGEI ͸ಈ͖Λॏࢹͨ͠ಛ௃
ͯ‫ͳͱ݈ؤ‬ΔՄೳੑ͕͋Δɽর߹ͷࡍ͸ɼࠨӈ্Լͷಈ͖
Ͱ͋Γɼ‫ב‬΍ίʔτͱ͍ͬͨਓ෺ྖҬͷมԽʹରͯ͠‫݈ؤ‬
ํ޲ͱ੩ࢭ෦෼ͷ‫ͭޒ‬ͷ‫ه‬ड़ࢠຖʹ૬ҧ౓Λ‫ٻ‬ΊΔɽ͜͜
ͳ܏޲ʹ͋Δɽਤ 1(d) ͸ GEI ΛՄࢹԽͨ͠ը૾ͱ GEnI
Ͱɼ‫ٻ‬Ίͨ‫ͭޒ‬ͷ૬ҧ౓ͷ಺ɼԼํ޲ͷ‫ه‬ड़ࢠ͸ࣝผʹ༗
͔Β࡞੒ͨ͠ϚεΫը૾Ͱ͋Δɽ
ޮͳ৘ใྔ͕গͳ͍ͨΊল͖ɼ࢒Γͷ࢛ͭͷ૬ҧ౓ͷॏΈ
GFI ͸ɼγϧΤοτը૾ྻʹର͢ΔΦϓςΟΧϧϑϩʔ
৔͔Βɼᮢ஋Ҏ্ͷಈ͖ͷ͋Δ෦෼Λप‫Ͱظ‬ฏ‫ۉ‬Խ͢Δ͜
෇͚૯࿨Λ࠷ऴతͳ૬ҧ౓ͱ͢Δɽ
ͱͰநग़͢Δา༰ಛ௃Ͱ͋ΔɽGFI ͸ΦϓςΟΧϧϑϩʔ
3. র߹ख๏
ۭ͔ؒΒநग़͞ΕΔͨΊɼγϧΤοτը૾ྻͷಈ͖ͷ෦෼
3.1 ϢʔΫϦου‫ͮ͘جʹ཭ڑ‬র߹ख๏
Λॏࢹͨ͠า༰ಛ௃ͱͳΔɽਤ 1(e) ͸ GFI ΛՄࢹԽͨ͠
র߹ख๏ͱͯ͠࠷΋୯७ͳํ๏͸ɼา༰ಛ௃ͷ֤࣍‫ݩ‬ͷ
ը૾Ͱ͋ΔɽγϧΤοτը૾ʹର͢ΔΦϓςΟΧϧϑϩʔ
ࠩͷࣗ৐࿨ͷฏํࠜɼଈͪɼา༰ಛ௃ϕΫτϧʹର͢ΔϢʔ
Λ‫͍ͯ͠ࢉܭ‬ΔͷͰɼಛ௃͕ਓ෺ͷྠֲ෇ۙʹͷΈ‫ݱ‬Εͯ
ΫϦου‫཭ڑ‬Λ૬ҧ౓ͱ͢Δํ๏Ͱ͋Δɽ͜͜Ͱ͸ɼา༰
͍Δ͜ͱ͕෼͔Δɽ
ը૾ྻ P ɼG ͔Βநग़͞ΕΔ D ࣍‫ݩ‬ͷา༰ಛ௃ϕΫτϧΛ
CGI ͸ɼ4 ෼ͷ 1 าߦप‫ظ‬ຖʹநग़ͨ͠ྠֲΛҐ૬ʹԠ
D
P
G
D
G
xP
i ∈ R (i = 1, . . . , N )ɼ‫ ͼٴ‬x j ∈ R (j = 1, . . . , N )
ٖͯ͡ࣅ৭Λ༻͍ͯϚοϐϯά͠ɼॏͶ߹Θͤͨา༰ಛ௃
Λর߹͢Δ͜ͱΛߟ͑Δɽ͜͜ͰɼN P ɼN G ͸ɼา༰ը
Ͱ͋ΔɽCGI ͸ྠֲΛಛ௃ͱͯ͠நग़͍ͯ͠Δ͜ͱ͔Βɼ
૾ྻ P ɼG ͔ΒͦΕͧΕநग़͞ΕΔา༰ಛ௃਺ (ຆͲͷಛ
෰૷มԽͳͲ֎తཁҼʹΑΔาߦঢ়‫گ‬ͷมԽʹ‫޲܏ͳ݈ؤ‬
௃ʹ͓͍ͯ͸प‫ظ‬ͷ਺) Ͱ͋Δɽ·ͣɼP ͷ i ൪໨ͷಛ௃
ʹ͋Δɽਤ 1(f) ͸ CGI ΛՄࢹԽͨ͠ը૾Ͱ͋Δɽ
G
xP
i ͱɼG ͷ j ൪໨ͷಛ௃ x j ͷ૬ҧ౓ Mi,j ΛҎԼͷΑ͏
ʹ‫͢ࢉܭ‬Δɽ
2.2 ϑϨʔϜؒࠩ෼ʹ‫ͮ͘ج‬า༰ಛ௃
าߦʹ͓͚Δಈతͳ෦෼Λநग़͢Δख๏ͱͯ͠ɼϑϨʔ
Ϝؒࠩ෼͕͋ΔɽϑϨʔϜؒࠩ෼͸ɼt ϑϨʔϜ໨ͷը૾
ͱ t + n ϑϨʔϜ໨ͷը૾ͷࠩ෼͕ᮢ஋Ҏ্ͷ఺ͱͯ͠ந
ग़͞ΕΔɽ
ϑϨʔϜؒࠩ෼Ͱ࡞੒ͨ͠ಈ఺ͷγʔέϯε͔Βಛ௃Λ
நग़͢Δख๏ͱͯ͠ɼCHLAC ͕͋ΔɽCHLAC Ͱ͸ɼϑ
ϨʔϜؒࠩ෼ʹͯநग़ͨ͠ಈ͖ྖҬʹ͓͚ΔͦΕͧΕͷϐ
ⓒ 2013 Information Processing Society of Japan
G
Mi,j = x P
i − xj
(1)
࠷ऴతͳ૬ҧ౓ M ͸ɼ֤ಛ௃ͷ૊ʹର͢Δ૬ҧ౓ Mi,j ͷ
࠷খ஋Λબ୒͢Δɽ
M = min Mi,j
i,j
(2)
3.2 ਖ਼४൑ผ෼ੳʹ‫ͮ͘ج‬র߹ख๏
ೝূੑೳͷ޲্ͷͨΊʹ͸ɼຊਓಉ࢜ͷ‫཭ڑ‬Λ࠷খԽ͠ɼ
3
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
ද 1
OU-ISIR Gait Database, The Treadmill Dataset
આ໌
ֶश
ςετ
σʔληοτ
(a) PCA
Dataset A
9 छྨͷ଎౓มԽ
14 ਓ
20 ਓ
Dataset B
࠷େ 32 छྨͷ෰૷มԽ
21 ਓ
47 ਓ
(b) LDA
ਤ 2 ֤ۭؒʹ͓͚Δ GEI ͷࢄ෍ਤ
ଞਓಉ࢜ͷ‫཭ڑ‬Λ࠷େԽ͢ΔΑ͏ͳಛ௃ۭؒΛߏங͢Δ͜
ͱ͕ॏཁͰ͋ΔɽͦͷΑ͏ͳۭؒ͸ઢ‫ܗ‬൑ผ෼ੳ (Linear
discriminant analysis, LDA) Λར༻͢Δ͜ͱͰಘΒΕΔɽ
ઢ‫ܗ‬൑ผ෼ੳͱ͸ɼΫϥε಺‫ڞ‬෼ࢄߦྻͱΫϥεؒ‫ڞ‬෼ࢄ
ߦྻͷҰൠԽ‫ݻ‬༗஋໰୊Λղ͘͜ͱͰɼΫϥε಺෼ࢄΛ࠷
ਤ 3
଎౓มԽ (্ஈ) ͱ෰૷มԽ (Լஈ) ͷྫ
খԽ͠ɼΫϥεؒ෼ࢄΛ࠷େԽ͢ΔΑ͏ͳಛ௃ۭؒͷ‫ج‬ఈ
ϕΫτϧΛಘΔख๏Ͱ͋Δɽ͜͜Ͱ͸ɼߴ࣍‫ݩ‬ͷา༰ಛ௃
Λओ੒෼෼ੳ (Principal component analysis, PCA) Λ༻
4. ࣮‫ݧ‬
͍ͯ௿࣍‫ݩ‬ͷۭؒʹࣹӨ͠ɼಘΒΕͨ௿࣍‫ݩ‬σʔλΛઢ‫ܗ‬
4.1 σʔληοτ
൑ผ෼ੳ (LDA) ͢Δɼਖ਼४൑ผ෼ੳΛ༻͍Δɽ
࣮‫ʹݧ‬͸ The OU-ISIR Gait Database, The Treadmill
·ͣɼֶशʹ༻͍Δಛ௃ͷฏ‫ۉ‬ϕΫτϧΛ༻͍ͯɼ֤ಛ
Dataset [15] ͷ಺ɼDataset AɼB Λ༻͍ͨɽͳ͓ɼDataset
௃͔Βࠩ͠Ҿ͍͓͖ͯɼओ੒෼෼ੳΛ༻͍ͯา༰ಛ௃Λ௿
A ͸ 2km/h ͔Β 10km/h ·Ͱ 1km/h ࠁΈͷ଎౓มԽΛ࣋
࣍‫ࣹʹۭؒݩ‬Ө͢Δɽ࣍ʹɼֶशʹ༻͍Δಛ௃਺ɼಛ௃ͷ
ͭ 34 ਓͷඃ‫ऀݧ‬ͷาߦγʔέϯε͔ΒͳΓɼDataset B ͸
࣍‫ݩ‬ΛͦΕͧΕߦɼྻͱͨ͠ߦྻʹରͯ͠ಛҟ஋෼ղΛద
࠷େ 32 छྨͷ෰૷มԽΛ࣋ͭ 68 ਓͷඃ‫ऀݧ‬ͷาߦγʔέ
༻͢Δ͜ͱͰɼ௿࣍‫ݩ‬ͷۭؒͷ‫ج‬ఈϕΫτϧΛ‫ٻ‬ΊΔɽਤ
ϯε͔ΒͳΔɽ
2(a) ͸ GEI ͷ PCA ۭؒʹ͓͚Δ্Ґ 2 ओ੒෼ͷࢄ෍ਤͰ
͋Δɽ
ද 1 ʹ࢖༻ͨ͠าߦγʔέϯεΛవΊΔɽ·ͨɼ଎౓ม
Խɼ෰૷มԽͷྫΛਤ 3 ʹࣔ͢ɽ
࣍ʹɼಘΒΕͨ௿࣍‫ݩ‬σʔλΛઢ‫ܗ‬൑ผ෼ੳ͢Δɽઢ‫ܗ‬
·ͨɼา༰ಛ௃ͷর߹ʹࡍͯ͠ɼσʔληοτΛΪϟϥ
൑ผ෼ੳͰ͸ɼԼ‫ه‬ͷ௨Γɼ௿࣍‫ݩ‬σʔλͷΫϥε಺‫ڞ‬෼
Ϧʔͱϓϩʔϒͷೋछྨʹ෼ׂ͢ΔɽΪϟϥϦʔ͸าߦ
ࢄ SW ͱΫϥεؒ‫ڞ‬෼ࢄ SB Λ‫͠ࢉܭ‬ɼೋͭͷ‫ڞ‬෼ࢄߦྻ
γʔέϯεͷొ࿥σʔλΛɼϓϩʔϒ͸าߦγʔέϯεͷ
ͷҰൠԽ‫ݻ‬༗஋໰୊Λղ͘ɽ
ೖྗσʔλΛҙຯ͢Δɽ͜͜Ͱ͸ɼDatasetA ͸ඃ‫ऀݧ‬ຖ
SW =
Nfi eatures
N
class
i=1
ʹಉঢ়‫گ‬ԼͰͷาߦγʔέϯε͕ 2 ύλʔϯ͋ΔͷͰɼͦ
(xi,j − mi )(xi,j − mi )T
(3)
j=1
ΕͧΕΛΪϟϥϦʔɼϓϩʔϒͱ͠ɼDataset B ͸ܰ૷ʹ
͋ͨΔ 1 छྨͷาߦγʔέϯεΛΪϟϥϦʔɼ࢒Γͷ࠷େ
31 छྨͷ෰૷ͷาߦγʔέϯεΛϓϩʔϒͱ͢Δɽ
SB =
N
class
Nfi eatures (mi − m)(mi − m)T
(4)
i=1
SW x = λSB x, x = 0
͜͜ͰɼNclass ͸ඃ‫਺ऀݧ‬ΛɼNfi eatures
(5)
4.2 ධՁई౓
า༰ೝূͷੑೳʹ͓͍ͯɼ1 ର 1 ೝূ (Vefirication) ͱ 1
ର N ೝূ (Identification) ͷೋͭͷγφϦΦΛߟྀ͢Δɽ
͸ i ൪໨ͷΫϥε
1 ର 1 ೝূͰ͸ɼর߹ʹΑΓಘΒΕͨ૬ҧ౓͕ᮢ஋ҎԼ
ͷಛ௃਺Λද͢ɽ·ͨɼxi,j ͸ i ൪໨ͷΫϥεͷ j ൪໨ͷ
Ͱ͋Ε͹ຊਓͱͯ͠ड͚ೖΕɼᮢ஋Λ௒͑Ε͹ଞਓͱ͠
ಛ௃Λɼmi ͸ i ൪໨ͷΫϥεͷฏ‫ۉ‬Λɼm ͸શಛ௃ͷฏ
ͯ‫ڋ‬൱͢ΔɽϓϩʔϒͱΪϟϥϦʔશͯͷ૊ʹରͯ͜͠
‫ۉ‬Λද͢ɽ
ͷΑ͏ͳ൑அΛߦ͍ɼ݁Ռͱͯ͠ຊਓ‫ڋ‬൱‫ޡ‬Γ཰ (False
ਤ 2(b) ͸ GEI ͸ LDA ۭؒʹ͓͚Δ্Ґ 2 ੒෼ͷࢄ෍ਤ
rejection rate, FRR) ͱଞਓडೖ‫ޡ‬Γ཰ (False acceptance
Ͱ͋ΔɽPCA ۭؒʹ͓͚Δࢄ෍ਤ (ਤ 2(a)) ʹൺ΂ɼLDA
rate, FAR) Λ‫ٻ‬ΊΔɽຊਓ‫ڋ‬൱‫ޡ‬Γ཰ͱଞਓडೖ‫ޡ‬Γ཰ͷ
ۭؒͷࢄ෍ਤ (ਤ 2(b)) ͸֤Ϋϥε಺Ͱີू͠ɼ֤Ϋϥε
૊Έ߹Θͤ͸ɼ૬ҧ౓ͷᮢ஋ΛมԽͤ͞Δ͜ͱͰมԽ͢Δ͜
ؒͰ෼ࢄ͍ͯ͠Δ͜ͱ͕෼͔Δɽ
ͱ͔ΒɼͦͷτϨʔυΦϑΛද͢ड৴ऀૢ࡞ಛੑ (Receiver
Operating Characteristics, ROC) ‫ۂ‬ઢ͕ಘΒΕΔɽ
ⓒ 2013 Information Processing Society of Japan
4
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
͜͜Ͱɼਤ 4(a), (b) ͸ϢʔΫϦου‫ͮ͘جʹ཭ڑ‬র߹ख๏
ʹର͢Δ ROC ‫ۂ‬ઢͱ CMC ‫ۂ‬ઢΛɼਤ 4(c), (d) ͸ਖ਼४൑
ผ෼ੳʹ‫ͮ͘ج‬র߹ख๏ʹର͢Δ ROC ‫ۂ‬ઢͱ CMC ‫ۂ‬ઢ
Λ͍ࣔͯ͠Δɽ·ͨɼਤ 4(e)ɼ(f) ͸ɼͦΕͧΕ౳Ձ‫ޡ‬Γ
཰ɼ1 Ґೝূ཰ͷ݁ՌΛ͍ࣔͯ͠Δɽ
෰૷มԽʹ͓͚Δา༰ೝূͷ৔߹ɼর߹͢ΔΪϟϥϦʔ
ͱϓϩʔϒͷਓ෺ྖҬͷγϧΤοτ͕େ͖͘มԽ͢Δͨ
ΊɼγϧΤοτʹ‫ͮ͘ج‬า༰ಛ௃ͷਫ਼౓͕௿͘ͳΔɽಛʹ
(a) ROC ‫ۂ‬ઢ (ϢʔΫϦου
(b) CMC ‫ۂ‬ઢ (ϢʔΫϦου
ਓ෺ྖҬͷγϧΤοτʹେ͖͘ґଘ͢Δ GEI ΍ FDF ͸ଞ
‫)཭ڑ‬
‫)཭ڑ‬
ͷา༰ಛ௃ͱൺ΂ͯೝূਫ਼౓͕௿͍ɽҰํɼΤϯτϩϐʔ
Λར༻ͨ͠ GEnI ΍ MGEI ͸γϧΤοτʹ‫ͮ͘ج‬า༰ಛ௃
ͷதͰ΋ಈ͖෦෼ʹண໨͍ͯ͠ΔͨΊɼߴ͍ਫ਼౓ͱͳͬͯ
͍Δ (ਤ 4(a), (b) ࢀর)ɽ
͔͠͠ͳ͕Βɼਖ਼४൑ผ෼ੳʹ‫ͮ͘ج‬র߹ͷ݁Ռ (ਤ 4(c),
(d)) Λ‫ݟ‬ΔͱɼGEI ΍ FDF ͕ GEnI ΍ MGEI ΑΓߴ͍ਫ਼
౓͕ग़͍ͯΔɽ͜Ε͸ GEnI ΍ MGEI ͕୯७ൺֱͷࡍʹਫ਼
౓͕޲্͢Δ΋ͷͰ͋ͬͯɼਖ਼४൑ผ෼ੳʹ‫ͮ͘ج‬র߹ख
(c) ROC ‫ۂ‬ઢ (ਖ਼४൑ผ෼ੳ)
(d) CMC ‫ۂ‬ઢ (ਖ਼४൑ผ෼ੳ)
๏ͷ৔߹ɼGEI ΍ FDF ͷํ͕ΑΓઢ‫ܗ‬൑ผ෼ੳ͕ޮՌత
ʹಇۭؒ͘Ͱ͋ͬͨ͜ͱ͕ཧ༝ͱͯ͠‫͛ڍ‬ΒΕΔɽ
·ͨਤ 4(e), (f) ͷ݁ՌΑΓɼਖ਼४൑ผ෼ੳʹ‫ͮ͘ج‬র߹
ख๏͕ϢʔΫϦου‫ͮ͘جʹ཭ڑ‬র߹ख๏ΑΓ༏Ε͍ͯΔ
͜ͱ͕෼͔ΔɽΑͬͯɼ࣍અҎ߱͸ਖ਼४൑ผ෼ੳΛ༻͍ͨ
݁ՌͷΈΛఏࣔ͢Δɽ
4.4 ଎౓มԽʹ͓͚ΔੑೳධՁ
(e) ౳Ձ‫ޡ‬Γ཰
ਤ 4
(f) 1 Ґೝূ཰
෰૷มԽʹ͓͚ΔੑೳධՁ
࣍ʹɼ଎౓มԽʹ͓͚Δา༰ೝূͷੑೳධՁΛߦ͏ɽ͜
͜Ͱɼਤ 5 ͸ 1 ର 1 ೝূͷ݁ՌΛɼਤ 6 ͸ 1 ର N ೝূͷ݁
ՌΛ͍ࣔͯ͠Δɽ·ͨɼਤ 7(a)ɼਤ 7(b) ͸ͦΕͧΕ౳Ձ
1 ର 1 ೝূʹ͓͍ͯ͸ɼຊਓ‫ڋ‬൱‫ޡ‬Γ཰ͱଞਓडೖ‫ޡ‬Γ
‫ޡ‬Γ཰ɼ1 Ґೝূ཰ͷ݁ՌΛ͍ࣔͯ͠Δɽ
཰ͷ྆ํ͕௿͍͜ͱ͕๬·͍͠ͱ‫͑ݴ‬Δɽ·ͨɼຊਓ‫ڋ‬൱
଎౓มԽʹ͓͚Δา༰ೝূͷ৔߹ɼর߹͢ΔΪϟϥϦʔ
‫ޡ‬Γ཰ͱଞਓडೖ‫ޡ‬Γ཰͕౳͍͠ͱ͖ɼͦͷ‫ޡ‬Γ཰Λ౳Ձ
ͱϓϩʔϒͷਓ෺ͷಈ͖͕େ͖͘มԽ͢ΔͨΊɼಈ͖෦෼
‫ޡ‬Γ཰ (Equal error rate, EER) ͱ‫ͼݺ‬ɼ1 ର 1 ೝূʹ͓͚
ʹண໨ͨ͠า༰ಛ௃ͷਫ਼౓΋௿͘ͳΔɽಛʹΦϓςΟΧϧ
Δࢦඪͱ͢Δɽ
ϑϩʔΛར༻ͨ͠ GFI ΍ GMD ͸ΑΓಈ͖෦෼Λར༻͢
1 ର N ೝূͰ͸ɼ͋Δೖྗಛ௃ͱ N ਓͷొ࿥ಛ௃ͷর
ΔͨΊɼଞͷา༰ಛ௃ͱൺ΂ͯ଎౓ࠩʹΑΔਫ਼౓ͷ௿Լ
߹ͰಘΒΕͨ N ‫ݸ‬ͷ૬ҧ౓Λ‫ʹݩ‬ϥϯΩϯάΛ࡞੒͢Δɽ
཰͕େ͖͍ɽҰํɼਓ෺ͷ‫ܗ‬ঢ়ʹେ͖͘ґଘ͢Δ GEI ΍
͜ͷͱ͖ɼ૬ҧ౓͕௿͍΄ͲϥϯΩϯάͰ্ҐͱͳΔɽ࣍
FDF ͸ɼಈ͖ಛ௃ͱൺֱ͢Δͱਫ਼౓ͷ௿Լ͕‫ݶ‬ఆతͱͳͬ
ʹɼ࡞੒ͨ͠ϥϯΩϯά͔Β֤ϓϩʔϒʹ‫·ؚ‬ΕΔຊਓಉ
͍ͯΔɽ
࢜ͷॱҐΛௐ΂ɼͦͷॱҐʹର͢Δྦྷੵ෼෍Λྦྷੵࣝผਫ਼
5. ߟ࡯
౓ಛੑ (Cumulative matching characteristics, CMC) ‫ۂ‬ઢ
ͱͯ͠ࢉग़͢ΔɽCMC ‫ۂ‬ઢʹ͓͍ͯ͸ɼྫ͑͹ɼ3 Ґʹର
༷ʑͳาߦঢ়‫گ‬Լʹ͓͚Δา༰ೝূख๏ͷੑೳධՁΛ
͢Δೝূ཰͕ 90%Ͱ͋Ε͹ɼຊਓಉ࢜ͷর߹ϖΞ͕ 3 ҐҎ
ߦͬͨ݁Ռɼর߹ख๏Ͱ͸ϢʔΫϦου‫ʹ཭ڑ‬ൺ΂ͯਖ਼४
಺ʹೖΔׂ߹͕ 90%Ͱ͋Δͱ͍͏͜ͱΛҙຯ͍ͯ͠Δɽ
൑ผ෼ੳΛ༻͍Δ΋ͷͷํ͕ੑೳ͕ྑ͍͜ͱ͕෼͔ͬͨɽ
า༰ೝূͷ໨ඪͱͯ͠ɼॱҐ͕௿͍ͱ͖ͷೝূ཰͕ߴ͍
͜Ε͸ਖ਼४൑ผ෼ੳʹΑΔΫϥεؒ෼ࢄΛ࠷େԽɼΫϥε
͜ͱ͕๬·͘͠ɼಛʹ 1 Ґೝূ཰͕ੑೳධՁʹ͓͍ͯྑ͘
಺෼ࢄΛ࠷খԽ͢Δۭؒͷߏங͕ཧ༝ͱͯ͠‫͛ڍ‬ΒΕΔɽ
༻͍ΒΕΔɽ
Ұํɼา༰ಛ௃ʹ͍ͭͯߟ͑ͯΈΔͱɼาߦঢ়‫ʹگ‬Αͬ
ͯਫ਼౓ͷେ͖ͳҧ͍͕‫ݟ‬ΒΕͨɽ͜͜Ͱ͸γϧΤοτʹ‫ج‬
4.3 ෰૷มԽʹ͓͚ΔੑೳධՁ
࠷ॳʹɼ෰૷มԽʹ͓͚Δา༰ೝূͷੑೳධՁΛߦ͏ɽ
ⓒ 2013 Information Processing Society of Japan
ͮ͘า༰ಛ௃Ͱ͋Δ GEI ͱಈ͖Λ໌ࣔతʹར༻ͨ͠า༰
ಛ௃Ͱ͋Δ GMD ΛྫʹऔΓ্͛ͯɼߟ࡯͢Δɽ
5
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
(a) 2 km/h vs. 4 km/h
(b) 2 km/h vs. 6 km/h
(a) 2 km/h vs. 4 km/h
(b) 2 km/h vs. 6 km/h
(c) 2 km/h vs. 8 km/h
(d) 2 km/h vs. 10 km/h
(c) 2 km/h vs. 8 km/h
(d) 2 km/h vs. 10 km/h
(e) 4 km/h vs. 6 km/h
(f) 4 km/h vs. 8 km/h
(e) 4 km/h vs. 6 km/h
(f) 4 km/h vs. 8 km/h
(g) 4 km/h vs. 10 km/h
(h) 6 km/h vs. 8 km/h
(g) 4 km/h vs. 10 km/h
(h) 6 km/h vs. 8 km/h
(i) 6 km/h vs. 10 km/h
(j) 8 km/h vs. 10 km/h
(i) 6 km/h vs. 10 km/h
(j) 8 km/h vs. 10 km/h
ਤ 5
଎౓มԽʹ͓͚Δ ROC ‫ۂ‬ઢʹΑΔੑೳධՁ
·ͣɼ෰૷มԽʹ͍ͭͯߟ࡯͢Δɽ෰૷͕มԽ͢Δ৔߹ɼ
ਤ 6 ଎౓มԽʹ͓͚Δ CMC ‫ۂ‬ઢʹΑΔੑೳධՁ
ͯ GEI Λ, ಈ͖Λ໌ࣔతʹར༻ͨ͠า༰ಛ௃ͱͯ͠ GMD
લड़ͷ௨Γɼਓ෺ͷγϧΤοτ‫ܗ‬ঢ়͕େ͖͘มԽ͢ΔͷͰ
Λ‫ͨ͠ࢉܭ‬΋ͷ͕ɼਤ 8(b)ɼ8(c)ɼ8(e)ɼ8(f) Ͱ͋Δɽ·ͨɽ
า༰ͷಈ͖ʹண໨͢Δ͜ͱ͕ॏཁͰ͋Δɽਤ 8(a)ɼਤ 8(d)
͜ΕΒͷը૾ؒͷࠩ෼Λ‫ͨ͠ࢉܭ‬΋ͷ͕ɼਤ 8(g)ɼ8(h) Ͱ
͸ಉҰਓ෺ʹΑΔҟͳΔ෰૷ͷาߦը૾Ͱ͋Δɽ͜ΕΒͷ
͋Δɽ͜ͷͱ͖ɼGEIɼGMD ʹ͓͚Δ੩ࢭ෦෼Ͱ͸ 2 ಛ
าߦγʔέϯεʹ͍ͭͯγϧΤοτʹ‫ͮ͘ج‬า༰ಛ௃ͱ͠
௃ؒʹେ͖ͳ͕ࠩ‫ݟ‬ΒΕΔ͕ɼGMD ʹ͓͚Δಈ͖෦෼Ͱ
ⓒ 2013 Information Processing Society of Japan
6
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
(a) ౳Ձ‫ޡ‬Γ཰
ਤ 7
(a)
(a)
(b)
(c)
(d)
(e)
(f)
(b) 1 Ґೝূ཰
଎౓มԽʹ͓͚Δ౳Ձ‫ޡ‬Γ཰ͱ 1 Ґೝূ཰
(b)
(c)
(g)
(h)
ਤ 9 ଎౓มԽʹ͓͚Δา༰ಛ௃ͷҧ͍ɽ্ஈ: ೖྗಛ௃ɼதஈ: ొ
࿥ಛ௃ɼԼஈ: ೖྗಛ௃ͱొ࿥ಛ௃ͷࠩɽ্ஈɾதஈʹ͓͍ͯ
͸ɼࠨ͔Βɼ‫ݪ‬ը૾ɼGEIɼGMD Λද͢ɽ·ͨɼ(g)ɼ(h) ͸ɼ
ೖྗͱొ࿥ͷ GEI ͱ GFI ͷࠩΛද͢ɽ͜͜Ͱɼ੺͸ೖྗಛ
(d)
(e)
(f)
௃ͷΈʹଘࡏ͢Δ෦෼ɼ྘͸ొ࿥ಛ௃ͷΈʹଘࡏ͢Δ෦෼Λ
ද͠ɼͦΕͧΕൺֱର৅ͷา༰ಛ௃ྖҬΛද͢ɽͭ·Γɼ੺ɼ
΋͘͠͸྘Ͱද͞Ε͍ͯΔྖҬ͸ൺֱ͢Δର৅ͱͷࠩΛࣔ͢ɽ
‫ʹٯ‬ແ࠼৭ͱͳ͍ͬͯΔྖҬ͸ɼ૒ํͰಛ௃͕౳͍͠෦෼Ͱ
͋Δɽ
έϯεʹ͍ͭͯɼγϧΤοτʹ‫ͮ͘ج‬า༰ಛ௃ͱͯ͠ GEI
Λɼಈ͖Λ໌ࣔతʹར༻ͨ͠า༰ಛ௃ͱͯ͠ GMD Λ‫ࢉܭ‬
(g)
ਤ 8
(h)
ͨ͠΋ͷ͕ɼਤ 9(b)ɼ9(c)ɼ9(e)ɼ9(f) Ͱ͋Δɽ·ͨɽ͜
෰૷มԽʹ͓͚Δา༰ಛ௃ͷҧ͍ɽ্ஈ: ೖྗಛ௃ɼதஈ: ొ
ΕΒͷը૾ؒͷࠩ෼Λ‫ͨ͠ࢉܭ‬΋ͷ͕ɼਤ 9(g)ɼ9(h) Ͱ͋
࿥ಛ௃ɼԼஈ: ೖྗಛ௃ͱొ࿥ಛ௃ͷࠩɽ্ஈɾதஈʹ͓͍ͯ
Δɽ͜ͷͱ͖ɼGMD ʹ͓͚Δಈ͖෦෼Ͱ͸ 2 ಛ௃ؒʹେ
͸ɼࠨ͔Βɼ‫ݪ‬ը૾ɼGEIɼGMD Λද͢ɽ·ͨɼ(g)ɼ(h) ͸ɼ
͖ͳ͕ࠩ‫ݟ‬ΒΕΔ͕ɼGEI ΍ GMD ʹ͓͚Δ੩ࢭ෦෼Ͱ͸
ೖྗͱొ࿥ͷ GEI ͱ GFI ͷࠩΛද͢ɽ͜͜Ͱɼ੺͸ೖྗಛ
͕ࠩখ͍͞ɽ͜ͷ݁Ռ͔Β଎౓มԽʹ͓͍ͯɼਓ෺ͷγϧ
௃ͷΈʹଘࡏ͢Δ෦෼ɼ྘͸ొ࿥ಛ௃ͷΈʹଘࡏ͢Δ෦෼Λ
ද͠ɼͦΕͧΕൺֱର৅ͷา༰ಛ௃ྖҬΛද͢ɽͭ·Γɼ੺ɼ
΋͘͠͸྘Ͱද͞Ε͍ͯΔྖҬ͸ൺֱ͢Δର৅ͱͷࠩΛࣔ͢ɽ
‫ʹٯ‬ແ࠼৭ͱͳ͍ͬͯΔྖҬ͸ɼ૒ํͰಛ௃͕౳͍͠෦෼Ͱ
͋Δɽ
Τοτ‫ܗ‬ঢ়ͷΈʹண໨ͨ͠ GEI ͕ GMD ʹൺ΂ͯྑ͍݁
Ռ͕ग़Δ͜ͱ͕༧ଌͰ͖Δɽ
Ҏ্͔ΒɼGEIɼGMDɼͦΕͧΕͷ௕ॴɼ୹ॴΛ֬ೝ͢
Δͱͱ΋ʹɼγϧΤοτʹ‫ͮ͘ج‬า༰ಛ௃ͱಈ͖Λ໌ࣔత
ʹར༻ͨ͠า༰ಛ௃ͷॏཁੑΛ֬ೝͨ͠ɽ
͸͕ࠩ΄ͱΜͲ‫ݟ‬ΒΕͳ͍ɽ͜ͷ݁Ռ͔Β෰૷มԽʹ͓͍
ͯɼಈ͖෦෼ʹண໨ͨ͠ GMD ͕ GEI ʹൺ΂ͯྑ͍݁Ռ
͕ग़Δ͜ͱ͕༧ଌͰ͖Δɽ
࣍ʹɼ଎౓มԽʹ͍ͭͯߟ࡯͢Δɽ଎౓͕มԽ͢Δ৔߹ɼ
6. ͓ΘΓʹ
ຊ࿦จͰ͸ɼ༷ʑͳঢ়‫گ‬Լʹ͓͚Δา༰ೝূͷੑೳධՁ
ʹ͍ͭͯड़΂ͨɽ࠷ॳʹɼධՁʹ༻͍Δา༰ಛ௃ʹ͍ͭ
લड़ͷ௨Γɼา༰ͷಈ͖͕େ͖͘มԽ͢ΔͷͰਓ෺ͷ‫ܗ‬ঢ়
ͯɼγϧΤοτʹ‫ͮ͘ج‬ಛ௃ɼϑϨʔϜؒࠩ෼ʹ‫ͮ͘ج‬ಛ
ಛ௃ʹண໨͢Δ͜ͱ͕ॏཁͰ͋Δɽਤ 9(a)ɼ9(d) ͸ಉҰਓ
௃ɼΦϓςΟΧϧϑϩʔʹ‫ͮ͘ج‬ಛ௃ͷࡾछྨʹ෼͚આ໌
෺ʹΑΔҟͳΔ଎౓ͷาߦը૾Ͱ͋Δɽ͜ΕΒͷาߦγʔ
ͨ͠ɽ࣍ʹɼর߹ख๏ͱͯ͠ɼϢʔΫϦου‫ͮ͘جʹ཭ڑ‬
ⓒ 2013 Information Processing Society of Japan
7
Vol.2013-CVIM-187 No.10
2013/5/30
৘ใॲཧֶձ‫ڀݚ‬ใࠂ
IPSJ SIG Technical Report
র߹ख๏ɼਖ਼४൑ผ෼ੳΛ༻͍ͨর߹ख๏ʹ͍ͭͯઆ໌͠
ͨɽ࠷‫ʹޙ‬ɼThe OU-ISIR Gait Database, The Treadmill
[12]
Dataset ʹΑΔ࣮‫ݧ‬Λߦ͍ɼ଎౓͕มԽ͢Δঢ়‫گ‬ɼ෰૷͕
มԽ͢Δঢ়‫ ͍͓ͯʹگ‬1 ର 1 ೝূɼ1 ର N ೝূͷੑೳධՁ
Λߦͬͨɽ݁Ռͱͯ͠ɼྫ͑͹ GEnIɼMGEI ΍ GMD ͕
෰૷͕มԽ͢Δঢ়‫͍͓ͯʹگ‬ਫ਼౓͕޲্͢Δ౳ɼ֤ঢ়‫ʹگ‬
[13]
͓͍ͯదͨ͠า༰ಛ௃Λબ୒͢Δ͜ͱ͕ɼา༰ೝূʹ͓͚
Δਫ਼౓޲্ʹͭͳ͕Δ͜ͱ͕෼͔ͬͨɽ·ͨɼ୯७ʹಛ௃
[14]
ؒͷϢʔΫϦου‫͔཭ڑ‬Β૬ҧ౓Λ‫ٻ‬ΊΔख๏ʹൺ΂ͯɼ
ਖ਼४൑ఆ෼ੳΛ༻͍֤ͯಛ௃ͷ‫཭ڑ‬Λ‫ٻ‬ΊΔख๏͕ɼา༰
ೝূͷর߹ख๏ͱͯ͠ద͍ͯ͠Δ͜ͱ͕෼͔ͬͨɽ
[15]
ࠓ‫ޙ‬ͷ՝୊ͱͯ͠ɼ(1) େ‫ن‬໛σʔλϕʔεʹର͢Δಈ
͖ʹ‫ͮ͘ج‬า༰ಛ௃ CHLACɼGMD ͷੑೳධՁɼ(2) ԰֎
Ͱͷา༰σʔλϕʔε (ྫ͑͹ɼHumanID Gait Database
[16] ͳͲ) ʹର͢ΔੑೳධՁɼ(3) ‫؍‬ଌํ޲΍าߦํ޲͕ม
Խ͢Δঢ়‫Ͱگ‬ͷੑೳධՁɼͷࡾ͕ͭ‫͛ڍ‬ΒΕΔɽ
[16]
Vision, pp. 257–270, October 2010.
T Kobayashi and N Otsu. “Action and simultaneous
multiple-person identification using cubic higher-order
local auto-correlation”. Proc of the 17th International
Conference on Pattern Recognition, Vol. 4, pp. 741–744,
August 2004.
K Bashir, T Xiang, S Gong, and Q Mary. “Gait representation using flow fields”. Proc. of the British Machine
Vision Conference 2009, September 2009.
H Iwama, M Okumura, Y Makihara, and Y Yagi. “The
OU-ISIR Gait Database Comprising the Large Population Dataset and Performance Evaluation of Gait Recognition”. IEEE Trans. on Information Forensics and Security, Vol. 7, pp. 1511–1521, October 2012.
Y. Makihara, H. Mannami, A. Tsuji, M.A. Hossain,
K. Sugiura, A. Mori, and Y. Yagi. The ou-isir gait
database comprising the treadmill dataset. IPSJ Trans.
on Computer Vision and Applications, Vol. 4, pp. 53–
62, Apr. 2012.
S Sarkar, P J Philips, Z Liu, I Robledo, P Grother, and
K Bowyer. IEEE Transactions on Pattern Analysis and
Machine Intelligence.
ँࣙ ຊ‫ڀݚ‬͸ JSPS ‫ج‬൫‫( ڀݚ‬S)21220003 ͷॿ੒Λड
͚ͨ΋ͷͰ͋Δɽ
ࢀߟจ‫ݙ‬
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
Uidai. http://uidai.gov.in/.
How biometrics could change security.
http:
//news.bbc.co.uk/2/hi/programmes/click_
online/7702065.stm, Oct. 2008.
R. Urtasun and P. Fua. “3D Tracking for Gait Characterization and Recognition”. In Automatic Face and
Gesture Recognition, 2004. Proceedings. Sixth IEEE International Conference on, Vol. 1, pp. 17–22, May. 2004.
S Nicholas and J Carter. “Towards pose invariant gait
reconstruction”. Proc. of the IEEE International Conference on Image Processing 2005, Vol. 3, pp. 261–264,
September. 2005.
J Han and B Bhanu. “Individual recognition using gait
energy image”. IEEE Trans. on Pattern Analysis and
Machine Intelligence, pp. 316–322, February 2006.
Y Makihara, R Sagawa, Y Mukaigawa, T Echigo, and
Y Yagi. “Gait recognition using a view transformation
model in the frequency domain”. Proc. of the 9th European Conf. on Computer Vision, pp. 151–163, May
2006.
Y Makihara and Y Yagi. “Silhouette Extraction based
on Iterative Spatio-temporal Local Color Transformation
and Graph-Cut Segmentation”. Proc. of the 19th Int.
Conf. on Pattern Recognition, pp. 1-4, Dec 2008.
K Bashir, T Xiang, and S Gong. “Gait recognition using
gait entropy image”. in Proc. 3rd Int. Conf. Imaging
for Crime Detection and Prevention,, pp. 1–6, December 2009.
Toby HW Lam, K. H. Cheung, and James NK Liu.
“Gait flow image: A silhouette-based gait representation
for human identification”. Pattern recognition, Vol. 44,
No. 4, pp. 973–987, April 2011.
K Bashir, T Xiang, and S Gong. “Gait recognition without subject cooperation”. Pattern Recognit. Letters,
Vol. 31, No. 13, pp. 2052–2060, October 2010.
C Wang, J Zhang, J Pu, X Yuan, and L Wang. “Chronogait image: a novel temporal template for gait recognition”. Proc. of the 11th European Conf. on Computer
ⓒ 2013 Information Processing Society of Japan
8
Fly UP