Analysis of gait recognition system developed in NI LabVIEW

1 Co esponding au ho : [email p o ec ed] Analysis o gai ecogni ion sys em de eloped in NI LabVIEW Dániel Salánki1 , Ko nél Sa ajcz 1Uni e si y o Deb ecen, Facul y o Enginee ing and Mecha onics, 4028 Deb ecen, Hunga y Abs ac . Nowadays, he wo ld o biome ic iden i ie s is imp o ing one o he mos signi ican ly in secu i y echnology. Wi hin he biome ic iden i ie s, he esea ch g oup is wo king wi h he gai ecogni ion speciali y. The esea ch g oup ealized a complex gai ecogni ion sys em in NI LabVIEW, ha can de ec mo e e e ence poin s simul aneously wi h a uni e sal came a and can sui a unc ion on he de ec ed poin s. Mo eo e , he p og am can compa e he unc ions o he e e ence cu e wi h he unc ions o he ac ual cu e and e alua e i he wo gai images a e he same o no . The p og am can sa e he ac ual gai da a and eload hem. The leg/ oo a io analyse p og am be o e he gai ecogni ion is designed o imp o e accu acy. The esea ch g oup es ed he p og am in di e en ways: walking only in socks in sho s o in ouse s, walking in slippe s, walking in he opposi e way. These ac s ha e a ec ed he gai ecogni ion’s accu acy, he lowes False Rejec ion Ra e (FRR) was in he case o walking in socks in sho s. 1 In oduc ion In oday's wo ld, I ind i impo an o p o ec ou alues wi h he mechanical and elec onic ools a ou disposal. As a mecha onics enginee s uden , I sough o ind a heme ha plays a decisi e ole in e e yday li e and is closely ela ed o mecha onics. Nume ous biome ic iden i ica ion me hods exis oday (e.g. ace, oice, palm, i is, e ina, inge p in au hen ica ion and DNA – Deoxy ibonucleic Acid – analysis), bu gai ecogni ion is less common. This is how we came up wi h he idea o wan ing o implemen a came a-based gai ecogni ion p og am using he Vision module o Na ional Ins umen s LabVIEW so wa e, which is capable o eal- ime image p ocessing o ecognize ha he gai image o a pe son passing by in on o he sys em is he same as ha o he p e iously s o ed pe son’s gai image. I his condi ion is me , he sys em gi es him / he pe mission. The gai ecogni ion is p eceded by a oo examina ion ha examines he leg/ oo a io o he pe son. I he es is passed, he ac ual gai ecogni ion begins. 2 Gai ecogni ion – Re e ences People can o en eel ha a amilia pe son is al eady ecognized a om hei walk. This common expe ience is he idea behind he ac ha dis ic ecogni ion sys ems can be used in secu i y echnology. You can easily make an image o a pe son's walk, e en in public places, wi hou e en ge ing obse ed by ha pe son. Walking is also in luenced by se e al ac o s: oo wea , soil, a igue, cu en s a e o mind o any inju y [1]. Walking can be de ined as a se ies o cyclic and coo dina ed mo emen s ha esul in human displacemen [1]. Gai is de ined as “a manne o mo ing on oo ; way o walking o unning” in Webs e 's New Collegia e Dic iona y. Howe e , human gai is mo e han ha : i is an idiosync a ic ea u e o a pe son ha is de e mined by, among o he hings, an indi idual's weigh , limb leng h, oo wea , and pos u e combined wi h cha ac e is ic mo ion [2]. Compa ed o physiological biome ic iden i ica ion ( inge p in au hen ica ion, i is and ace iden i ica ion, ea lobe geome y iden i ica ion), he gai ecogni ion has he ad an age o iden i ying pe sons wi h low esolu ion o hidden came a sho s [3]. Physiological biome ic me hods canno eliably de ec non-coope a ing indi iduals, especially no a dis ances, unde he changing en i onmen al condi ions o he eal wo ld [4]. Howe e , he accu acy o walking ecogni ion can be g ea ly in luenced by se e al ac o s, such as he iewing angle ha changes he isual a ibu es [5]. 3 Leg/ oo a io analysis The essence o his analysis is ha we de e mine he a io o he leg's and oo ’s leng hs wi h a p og am de eloped in LabVIEW. We i s measu ed he a io o one o ou legs, which anged be ween 2.42 and 2.43, so we selec ed hese wo numbe s as limi alues. So, i he a io o he leg and oo measu ed in pixels is be ween hese wo numbe s, he p og am will gi e a g een signal. Fig. 1 shows a measu emen in ouse s and in slippe s. The p opo ional alue (2.26314) did no co espond o he se limi s, so he g een ligh did no lash. Fig. 2 shows ano he measu emen . In his case, he pe son was si ing in socks, wi hou any oo wea . In his ANNALS OF THE UNIVERSITY OF ORADEA Fascicle o Managemen and Technological Enginee ing ISSUE #2, AUGUST 2018, h p://www.im uo adea. o/auo. m e/ 65 case, he p opo ional alue was adequa e, and he g een ligh lashed. In sho s o wi hou socks he g een ligh also lashed. I can be s a ed ha he clo hes ha e minimal in luence on he esul s, bu he oo wea has an impo an ole in his measu emen . Fig. 1. Measu ing in slippe s. Fig. 2. Measu ing wi hou any oo wea . 4 The p esen a ion o gai ecogni ion so wa e We a ached a poin o he knee and ano he o he leg o he pe son, abou 35 cen ime es om he g ound. The backg ound o he poin s is whi e, he poin i sel is black. I is impo an o ha e a ligh o whi e backg ound o a oid alse poin de ec ion, and o ha e you oo close o he came a whe e he e e ence poin s a e loca ed. The measu emen s we e made wi h a i icial ligh , wi h no mal oom ligh ing, beside he da kened d ape y a oiding sunligh , c ea ing app oxima ely he same ligh ing condi ions o each measu emen . Fig. 3 shows he on panel while he p og am is wo king. On he igh side o he igu e, he coo dina es o he de ec ed poin s will be s o ed in blocks a e unning. X1 and Y1 a e alid o he i s poin (sui ed on he knee), while X2 and Y2 a e alid o he second poin . In he middle i is he ac ual image, he wo ed ec angles ame he e e ence poin s. On he second and hi d abs, he gai image is p ocessed. On he second ab he ac ual gai image is p ocessed, while on he hi d ab he sa ed and eloaded gai image is p ocessed. I he ac ual gai image is like he sa ed one, he g een ligh below he ac ual image will lash. The p og am sui s a unc ion on he cu es d awn by he wo e e ence poin s. You can choose be ween ou unc ion models (Linea , Polynomial, Exponen ial, Powe ) and you can choose he o de o he unc ion oo. On Fig. 4. a) and b) he blue poin s ep esen he de ec ed poin s and he ed cu es a e he unc ions sui ed on hem. Acco ding o he es s, he mos accu a e i is he ou h-deg ee (biquad a ic) polynomial unc ion (Fig 4. a)). Howe e , o biquad a ic polynomials, he hi d and ou h coe icien s ake e y small alues (up o 10-10), and he second-o de (quad a ic) unc ion also ollows well he cu e d awn by he poin s (Fig 4. b)), so he use o he ou h-deg ee polynomials becomes unnecessa y. Wo king wi h quad a ic unc ion is simple han wo king wi h biquad a ic ones. Fig. 4. a) Sui ing a ou h-deg ee polynomial unc ion on he gai cu e. Fig. 4. b) Sui ing a second-deg ee polynomial unc ion on he gai cu e. ANNALS OF THE UNIVERSITY OF ORADEA Fascicle o Managemen and Technological Enginee ing ISSUE #2, AUGUST 2018, h p://www.im uo adea. o/auo. m e/ 66 Fig. 3. The on panel o he p og am while wo king. 5 Gai ecogni ion esul s in di e en condi ions Fig. 5. shows he coe icien s o he unc ions sui ed on he cu e d awn by he i s (blue columns) and he second poin (g een columns). Du ing his se ies o measu emen s, he es ed pe son wo e sho s and socks wi hou any shoes o slippe s, and he i s s ep was made wi h he igh leg ( he e e ence poin s we e also ixed on he igh leg). Acco ding o he esul s, in 8 cases ou o 10 he sys em ecognized he es ed pe son. The unsuccess ul a emp s a e w i en wi h ed cha ac e s. Fig. 5. Resul s in sho s, no oo wea . On Fig. 6. you can see he coe icien s o he unc ions ecei ed om measu emen s made in sho s wea ing slippe s. The esul s a e e y simila o he i s measu emen , he False Rejec ion Ra e (FRR) inc eased a bi : he es ed pe son was ecognized 7 imes ou o 10 measu emen s. Fig. 6. Resul s in sho s, in slippe s. The unsuccess ul a emp s a e w i en wi h ed cha ac e s. Fig. 7. shows he da a ecei ed om measu emen s made in ouse s, wi hou shoes. The FRR inc eased o 50%, so only hal o he measu emen s we e success ul. These esul s can be explained wi h he ac ha he e e ence poin s ixed on he ouse s we e mo ing wi h he clo h ( he ouse s a e mo ing on he leg while walking), while ixed di ec ly on he es ed pe son’s knee and leg, hey ollowed exac ly he mo emen o he knee and he leg. Fig. 7. Resul s in ouse s, no oo wea . The unsuccess ul a emp s a e w i en wi h ed cha ac e s. ANNALS OF THE UNIVERSITY OF ORADEA Fascicle o Managemen and Technological Enginee ing ISSUE #2, AUGUST 2018, h p://www.im uo adea. o/auo. m e/ 67 Fig. 8. shows he esul s om a se ies o measu emen s made in sho s, wea ing no shoes, he e e ence poin s ixed on he le leg, and going om he igh o he came a o he le . The p e ious measu emen s we e made om he le o he came a o he igh . Wi h hese a emp s, he FRR became 90%, only one ime ou o 10 was he es pe son ecognized. This esul was caused by he ac ha he so wa e was calib a ed o s a om he le o he came a. The cu es d awn by he e e ence poin s ixed on he le o on he igh leg a e di e en . Fig. 8. Resul s in sho s, walking in he opposi e di ec ion. The unsuccess ul a emp s a e w i en wi h ed cha ac e s. The measu emen s p esen ed abo e we e made wi h s a ing he walking wi h he igh leg. We ha e made some measu emen s wi h he same condi ions as in he i s measu emen se ies ( e e ence poin s on igh leg, wea ing sho and no oo wea ), bu he i s s ep was made wi h he le leg. The esul s a e shown on Fig. 9. Because he coe icien s we e e y di e en om he ini ial alues, we made only h ee a emp s, ce ainly wi hou any success. The cu es d awn by he wo e e ence poin s, ha a e shown on Fig. 10. a) and b) we e a om he ini ial cu es because o he cyclical shi and because he came a can iew only a ew s eps. So, i is no mal ha he p og am didn’ make any ecogni ion in his measu emen . Fig. 9. Unsuccess ul esul s s a ing wi h he le leg. Fig. 10. a). Cu e d awn by he second e e ence poin s a ing he walking wi h le leg Fig. 10. b). Cu e d awn by he second e e ence poin s a ing he walking wi h igh leg 6 Conclusion Wi h hese es s, we ha e showed how ou gai ecogni ion sys em is wo king in di e en si ua ions. Acco ding o he esul , he oo wea has he leas in luence on he sys em, bu wea ing ouse s, o walking in he opposi e di ec ion can ha e se ious impac on he success ulness o he es . I is also impo an o s a he measu emen wi h he igh leg, because o he wise he cu es will be a di e en om he e e ence cu es, and he p og am will no ecognize e en he pe son wi h pe mission. Acknowledgemen The wo k/publica ion is suppo ed by he EFOP-3.6.1-16- 2016-00022 p ojec . The p ojec is co- inanced by he Eu opean Union and he Eu opean Social Fund. Re e ences 1. Je ey E. Boyd, James J. Li le: Biome ic Gai Recogni ion, 19-42 (Sp inge - Ve lag Be lin Heidelbe g, 2005) 2. Lee, Lily, W. E ic L. G imson. Gai analysis o ecogni ion and classi ica ion. Au oma ic Face and Ges u e Recogni ion, 2002. P oceedings. Fi h IEEE In e na ional Con e ence on (pp. 155-162). IEEE, 2002. 3. Ma k S. Nixon, T. N. Tan, R. Chellappa: Human Iden i ica ion Based on Gai , (Sp inge Science + Business Media, Inc., 2006) 4. Ju Han, Bi Bhanu: Indi idual Recogni ion Using Ga e Ene gy Image, 28.2: 316-322, (IEEE ansac ions on pa e n analysis and machine in elligence, 2006) 5. Wei Zeng, Cong Wang: Neu ocompu ing - View- in a ian gai ecogni ion ia de e minis ic lea ning, 175: 324-335 (2016) ANNALS OF THE UNIVERSITY OF ORADEA Fascicle o Managemen and Technological Enginee ing ISSUE #2, AUGUST 2018, h p://www.im uo adea. o/auo. m e/ 68