• Aucun résultat trouvé

Slicing the 3D space into planes for the fast interpretation of human motion

N/A
N/A
Protected

Academic year: 2021

Partager "Slicing the 3D space into planes for the fast interpretation of human motion"

Copied!
1
0
0

Texte intégral

(1)of a pulsed laser beam and the reception of its echo; this time is used to estimate the distance.. Slicing theUTM-30LX. 3D space into planes F igure 1. Hokuyo for the fast interpretation of human motion Sébastien Piérard. Marc Van Droogenbroeck. Samir Azrour. Sebastien.Pierard@ulg.ac.be. Samir.Azrour@ulg.ac.be. M.VanDroogenbroeck@ulg.ac.be. !"# $"%&' %(") !*++# !,# $"%&' %(") !*-.# !(# %(") !*/*# INTELSIG, Montefiore Institute, University of $"%&' Liège, Belgium. Laser Scanners: a suitable technology for markerless motion analysis The UTMRange has a scanning range from !135° to +135° in steps of 0.25° (0° is in the human front of the device) with a resolution of 1 mm, and one scan is completed in 0.025 s (40 Hz). The raw scan data provided by the device contains 1,081 distance-points expressed in millimeters and represented using • Single-row range laser scanners measure the distance between themselves and the closest element, in a set of predefined directions that are usually located in a 18 bits with a total of 3,243 bytes per scan. plane. Therefore, they can be used to analyse the geometry of a thin slice of the observed scene. The UTM attached to !*++# an Intel Core 2,signal 1.66 (e.g. GHz, GB345623&%7% RAM portable a USB !0#was 1&%2 345623&%7% !&#a 1pulse). 1&%2 !*-.# !8# 1&%2 345623&%7% !*/*#at • They derive the distance measures from the time-of-flight of an®infrared A large numbercomputer of precise using distance measures can be taken 2.0directions port for scanned registering data.orThe UTM was controlled proprietary (open) based on high frequency (e.g. 274 at 60Hz, 16440 measures per second,using in the acase of the sensor BEAprotocol LZR-U901). • The suitability of rangemessages. laser scanners on the speedof of the observed and is in high for human motion [1].ms An depends interesting feature UTM iselements, the inclusion of general an internal reference time (1 resolution) in the raw data provided to avoid the effect of the delays originated in the USB communication with the computer. The wide variety of potential applications for human motion analysis with range laser scanners !9# to $"%&' %(") the !.::#gait parameters !3# $"%&' %(") !.:;#the scans !7# $"%&' %(") !.:+# The measurement procedure used obtain from acquired with the LIDAR is stated in [39]. Figure 2 shows a representation of a typical scanning procedure during a walking experiment. The LIDAR was placed at a height of 100 mm above the ground (approximately at ankle height) with the scanning plane parallel to the floor to obtain the maximum information from the legs without detecting the shoes/feet. Figure 3 shows an example of the raw data points obtained in the case of two isolated legs in front of the LIDAR. This raw data is segmented and used to fit two circles that define the instantaneous position of the legs of the subject while walking [39] and then the parameters of the gait are computed. '(. '". !(. !". &(. &". (. ". !%& !%" ''& '"% !*) %$* *') *'% !$) *"( #!% !#% !*% '"! ')' )'"#(& !)# %!% !*"'%" %"% *'! ')# (&! '!' !'$ !#& (&% '&* ''% #%% !*' '#*#"$ *#"%"* )!* (&$'*% #%$ '($ !$('(! '#) #!) $"$ #*" ()% ##*$'! '** (') '$*#"( (%" ''" #%# !#! !%! !)' '&" '"& )## (&( !*& $") ')"')!#!( !(* $)! '*) '(# #%* '(" !$! #*& ''( !$$ $'& (&# '!% ((( '%& )"% #") !%* !(& '&! !)*!$% #%! ''' '"$ !#) ()) #"! $)& !)& #"# !(% '!* '*$ #!# !$" !(# ('$ '')*"$ ()* ''#!$' *'# '"" $'' '!! ''! (&)'!" $"( !'% '!& '!$ #"" !*! ')& !)!!)( !*( !)" '"' #%( #%' '!) '"*(&' !$* '(' '&% #!$#!' '(( '"# )%) !$# *'' '!# !%% #%)')( !$& !#" !(" $'# ')* #"' '%! #'" !%' '(& ('% !*# !)%. &". !". '". #". (". Figure 1: BEA LZR (real size). '(. '". !(. &"". !". *" %" )". &(. $" (" #". &". '" !" &". (. $". ". !". &". !". '". #". (". #". $". %". &"". &!". &#". &$". $". 1&%2 345623&%7% !.:;# 1&%2 345623&%7% !># 1&%2 345623&%7% !.:+# Figure 2: By analysing a thin horizontal slice !<# of the scene, it is for !.::# Figure 3: Analysing the!=# behaviour of people and their interactions is also Figure 4: When a vertical slice of the scene is observed with range laser example possiblesemantic to track more 150 Once peoplewesimultaneously with aresults few (Fig.a), possible, range laser scanners in a doorframe, machine learning techniques can recognize Figure 3. Learning scene by than tracking. obtained the tracking these as trajectories werescanners clustered can into be used to find the temporally F igure 2. Representation of a typical measurement setand with a representation of the scan plane. piggybacking and tailgating scenarios [4] different crowd via an online unsupervised learning (Fig.b). Then, we could learn scene knowledge: Dynamic properties: density sensors, andflows to analyse crowd flows in public places [2]. coherent (1) groups of people to track these groups globally [3]. distribution (Fig.c) and velocity distribution (Fig.d) of the crowd flow. (2) Static properties: walk paths and sinks/sources (Fig.f). Note that the arrows in Fig.d show the principal orientation at each position. Please see texts for more details.  ?. ?. . .  at x(𝑡): become a crowd flow and it could be seen as a type of activvariables must vanish  !?# $"%&' %(") !..+# ⊤ !)# $"%&' %(") !.;.# ity. Secondly, we should extract the knowledge of the scene ⊤ ⊤ 𝑝𝑛 (x(𝑡), 𝑡) = (b1 (𝑡)x(𝑡))(b2 (𝑡)x(𝑡))...(b𝑛 (𝑡)x(𝑡)) = 0. Ek from these clusters. Scene knowledge should contain two (1) parts: dynamic properties (e.g., information of crowd flow) 9NVk +F This homogeneous polynomial can be written as a linear and static prosperities (e.g., walk paths and sinks/sources). %k 𝐼 combination of all the monomials of degree 𝑛 in x, x = z These scene knowledge can provide a great help to the y 𝑛1 + 𝑥𝑛1 1 𝑥𝑛2 2 ...𝑥𝑛𝐷𝐷 with 𝑛2 + ...𝑛𝐷 = 𝑛, as tracking and the overall pipeline is illustrated in Fig.3. In x . ∑ this section, we will provide *SQ

(2) m the

(3) m details about these items. !Kbc_JLcNMm<\\Zm 6\MNW m >N[b\_m :WJLNZN[c m J[Mm5JbN_m <J[QNm𝑝 >LJ[[N_m<NL\_MS[Qbm (x, 𝑡) = &e_S[Qm𝑒*SNWMm @_SJWm (𝑡)𝑥𝑛1 ...𝑥𝑛𝐷 = e(𝑡)⊤ 𝜇 (x) = 0,  ?. ?. ?. ?.  ?.  < ?. !6# $"%&' %(") !.@:#.

(4) ?. ?. ?. ?. '?. ?. < ?.  ?. UTM. 𝑛1 ,...,𝑛𝐷 𝑛 1 𝐷 Figure 5: Human motion analysis with range laser scanners is𝑛also a Figure 6: The gait characterization is also important in neurology, and (2) @!#5'm 0m cornerstone for some medical assessments. As mobility plays an important >C m omvv[H`K† em[kX†oExZv† HKx€KKk† xZmvK†vKkvmsv† JKxE[`v† [k† CD

(5) † @A+m measuring various gait where 𝑒𝐼 (𝑡) ∈ ℝ represents the coefficient of characteristics the monomial with a range laser scanner has also 2k`‚†xZK†SEx†mM†E††‚KEsv†m`J†MKeE`K†oEsx[I[oEkx†€Ev†Kr}[ooKJ† roleOnline in geriatrics, measuring the walking speed of elderly ;Ck people 𝐼H<A0kwith a 𝐷 3.1. clustering €[xZ† E† /9;† )m^}‚m† =9( /A† €Z[IZ† €Ev† o`EIKJ† Ex† E† x . The map 𝜇𝑛 : ℝbeen →proposed ℝ𝑀𝑛 (𝐷) [6]. is known as Veronese k CHWUTYQk <[cHWk <[cHWk @^YTYQk CHWUTYQk ;cHYLTYQk ;TccTYQk range laser scanner placed in their flats has been proposed [5]. IKkxsE`† om[kx† €[xZ[k† xZK† Smms† '[X}sK† † [k† xZK† e[JJ`K

(6) † <ZK† <RO^Ok A\k ,HJUk .[gYk  ' '&'

(7) ' ' "'. '  '%'. Figure 7: Some preliminary results suggest that range laser scanners could also be used to recognize people, with machine learning techniques, based on their gait [7], which has potential applications in intelligent houses, as well as in security.. map [10] of degree 𝑛, where 𝐼 is chosen in the degree

(8) m 

(9) m 

(10) m 

(11) m 

(12) m 

(13) m 

(14) m -m Problem Formulation and Overview: lexicographic order and 𝑀 (𝐷) is the total number of in

(15) m m 

(16) m 

(17) m 

(18) m 

(19) m 𝑛 

(20) m 

(21) m Going further in the analysis and interpretation: the project GAIMS (gait analysis in multiple sclerosis) 

(22) m 

(23) m m 

(24) m 

(25) m 

(26) m 

(27) m 

(28) m dependent monomials. !..+# !A# 1&%2 345623&%7% !.;.# !'# 1&%2 345623&%7% !.@:# At time 𝑡, person 𝑖 is represented by x!5# (𝑡)1&%2 = m345623&%7% 

(29) m 

(30) m 

(31) m 

(32) m 

(33) m 

(34) m 

(35) m to

(36) m the polynomial equation (2), we can perform 

(37) m 

(38) m 

(39) m 

(40) m (𝑥𝑖 , 𝑦𝑖 , 𝑣𝑖𝑥 , 𝑣𝑖𝑦 ), where (𝑥 , 𝑦 ) is the position, (𝑣 , 𝑣 ) the m 

(41) m 

(42) mThank m 

(43) m 

(44) m 

(45) m 

(46) m 

(47) m 

(48) m 

(49) m online hyperplane clustering by on the polyvelocity. With the help of the trackers, we obtain the 

(50)  m the B79C *C 

(51)  m$"%&' %(") ")0operating ,&%2 345623&%7% 7) 5"'2 68 23& &D5&'7?&)2 !(4(>&% *++ 26 .@:# 

(52)  m 

(53)  m 

(54)  m 

(55)  m 

(56)  m 

(57) mcoefficients 

(58) m 

(59) me(𝑡)

(60) m 

(61) m nomial rather

(62) mthan on the normal vectors 𝑁 In trajectories (𝑡)} , awhere 𝐿 (𝑡) = {x (𝑡)project, : 𝑡 = 

(63) m 2011, we{𝐿started multidisciplinary named GAIMS, to analyse {b𝑗 (𝑡)}𝑛𝑖=1 . That is because e(𝑡) does not depend on which 1, ..., 𝑇 }. We should cluster these trajectories into 𝑛 clusand interpret the gait of people with multiple sclerosis (MS). Measuring 𝑛 * &)'&* hyperplane the x(𝑡) belong to. Therefore, at each time 𝑡, we ters {𝑆𝑗 (𝑡)}𝑗=1 at a specific time 𝑡. In order to dynami;[kIK† E† I`[k[IE``‚† E`[JExKJ† sKMKsKkIK† v‚vxKe† Mms† eKEv}s[kX† their gaittheischange important most of them have walking impairments ˆe(𝑡) of (7'(>&F seek to find an estimate e(𝑡) that minimizes cally reflect of scenebecause information, all the clus7% ">%6 ?"'=&0 E723 "kmx†%4?,6> 2'7")9>&F %AG"'& 6' " %7DC L) 8"(2F %&K&'"> 27?&% 6)>4 6)& 6' 2E6 2'"(=% "'& 0&2&(2&0F XE[x† €Ev† EE[`EH`K† €[xZ[k†!" xZK† SExv† xZK† mk`‚† IZEkIK† xm† mHxE[k† E`}Kv† Mms† ImeoEs[kX† xZK† sKv}`xv† mM† xZK† KoKs[eKkx† tering should online. Inspired by [26], weand consider since the be early stages of ('6%%#F theVidal disease, perceive it as the most disabling 𝑡EWKs† 𝑁 ∑ ∑ xm† €Ev† xm† oKsMmse† Ek† EvvKvveKkx† HKMmsK† EkJ† xZK† xs[E`† E37>& ?&"%G'&?&)2% "'& '&5'&%&)2&0 ,4⊤ 062%C 2 "% 72 3"55&)%F 86' &D"?5>&F E723 (>G%2&' "C 1 each cluster 𝑆 (𝑡) as a moving hyperplane. Thus, we can 𝑗 €Ev† vxEsxKJ

(64) † <EH`K† ,† vZm€v† xZK† sKv}`xv† mM† v[† <=(† xKvxv† 𝑓 (e(𝑡)) = (e(𝜅) 𝜇𝑛 (x𝑖 (𝜅))) . (3) symptom [8]. Thus, gait plays an importanteKEv}sKJ† role }v[kX† forE† showing there is no ;@† EkJ† 𝑁 xZK† E<=(† EooEsEx}v

(65) † model a union of 𝑛 hyperplane in ℝ , where 𝑆 (𝑡) = {x ∈<ZK† Rsvx† x€m† xKvxv†vxmo€ExIZ† 𝑖=1 €KsK† oKsMmseKJ† xZK† JE‚†𝜅=1 xZK† KoKs[eKkx† ⊤ B79C *!0# %36E% 23& 7)727"> %72G"276) "2 J2 (4(>& *-. !B79C *!&##F (>G%2&'% % ")0 $ 9&2 K&'4 (>6%&F 𝐷 evidence of disease activity [9], as well as for demonstrating the efficacy of ,&97))7)9 68 237% ℝ : b𝑗 (𝑡)x = 0}, 𝑗 = 1, ..., 𝑛, where b(𝑡) ∈ ℝ , as thevxEsxKJ † xKvxv†  † Ex† xZK† `Evx† JE‚† mM† xZK† xs[E` † EkJ† 23& xKvxv†  † gradient descent, we obtain the folmkK† €KK^†By EWKs†using xZK† RK`J†normalized xs[E`† €Ev† vxmooKJ

(66) † <ZK† xKvxv† €KsK† zero set of a polynomial with time5"'2 varying coefficients using therapies. Figure 8: There are 2,500,000 Figure 10: the most"'& AG72& Figure 11: Gait helps to show 68 23& &D5&'7?&)2C H>G%2&' !xZK†identifier 3"% 6)& 5&'%6) ")0 (>G%2&'% ,G2 23&'& 7% 9:)6The<67) 68 course 23&%&of (>G%2&'% "% Gait 23&is%2"2&% Figure disease oKsMmseKJ† EkJ† KE`}ExKJ† KEIx`‚† vEeK† €E‚† Ev† JKvIs[HKJ† lowing recursive normalized gradient descent. Then the hyperplane normals[k† CD

(67) † 9Kv}`xv† vZm€† xZEx† xZK† oEsx[I[oEkx† Imeo`KxKJ† E``† <=(† people with MS in the world. important domain for23"2 MS. (4(>& the absence of disease activity MS0G& and the EDSS score. "#$ 23'&& 5&65>& &"(3C I3& 0788&'&)2 2'"(&0 >7)&% '&5'&%&)2 0788&'&)2 26 23&7' 3&"07)9 07'&(276)%C B'6? 26 xKvxv† €[xZ[k† xZK† sEkXK† mM† 

(68) v† xm† 3

(69) v† eKEv}sKJ† €[xZ† ;@† are estimated from the derivatives of the new polynomialeKEk†  vxJ † EkJ† HKx€KKk† 

(70) v† EkJ† 

(71) v† eKEv}sKJ† €[xZ† v(𝑡) ˆe(𝑡 + 1)23"2 = ˆe(𝑡) cos(∥v(𝑡)∥) + sin(∥v(𝑡)∥), (4) (4(>& */MF (>G%2&' $ 7% 5"'27">>4 6' 262">>4 6((>G0&0 ,4 (>G%2&' xZK† E<=(† EooEsEx}v

(72) † <ZK† J[MMKsKkIK† mM† 

(73) v† HKx€KKk† ;@† 86>>6E&0 0G'7)9 23& "55'6D7?"2& 2'"<&(26'7&% 23& (>G%2&'% at each trajectory. Lastly, the trajectories are grouped by ∥v(𝑡)∥ eKEv}sKeKkxv†EkJ† eKEv}sKeKkxv† Msme† E<=(† EsK† kmx† sK`KEkx† clustering associatedthe normal vectors. Mms†E†J[OKsKkx†I`Evv[RIEx[mk

(74) †1m†v[Xk[RIEkx†J[OKsKkIK†HKx€KKk† GAIMStheir measures trajectories of the%&AG&)(&C lower limbs extremities and de23& 07%5>"4&0 J% 23& )G?,&' 68 5&65>& " (>G%2&' %C J% ") &D"?5>&F 7) (4(>& */* !B79C *!8## 6)>4 6)& 2'"(= 7% where the negative normalized gradient is7) computed as xZK† xKvxv†oKsMmseKJ† HKMmsK†EkJ† EWKs† xZK† RK`J† xs[E`† €KsK† Mm}kJ

(75) † ;xEkJEsJ† JK[Ex[mk† Mms†E<=(† sKv}`xv† [v†`m€Ks† xZEk† Mms†xZK† ;@† Algorithm Detail: rives many gait characteristics from them The system GAIMS is non7)('&"%&%F 23& [1]. 6((>G%76)% ,&2E&&) 23&? 72 K7'2G">>4 0&2&(2&0 7) (>G%2&' $ !23& (>G%2&' =&&5% 2'"(= 68 2E6#F E37>& ⊤ eKEv}sKeKkxv† €Z[IZ†om[kxv†xm† v}H]KIx[K†}vEXK†mM† xZK†ˆ;@†?"=& H‚† ˆ v(𝑡) = −𝛽(𝐼 − e (𝑡) e (𝑡)) 𝑀𝑛 (𝐷) xZK† Z}eEk†that xKvxKs

(76) † =v[kX† mk`‚† xZK† x[eK† Mms† Imeo`Kx[kX† €E`^[kX† intrusive, in x(𝑡) the insense it is contactless and its usage does not Given a point one ofthat hyperplane 𝑆 (𝑡), there is 𝑗 ∑ 7?56%%7,>& 26 0&2&(2 68 " 9'6G5C B6' (>G%2&'% 86G' ?&"%G'&?&)2% "'& 0&2&(2&0 86' (>G%2&' % !23& (>G%2&' =&&5% 𝑁 JK`E‚† xZKsK† EkJ†">> €E`^[kX†23& HEI^ † vm†2'"(=% KI`}J[kX† E``† Msme† vxEkJ[kX† ⊤ ⊤ (5) ˆ (e clin(𝑡)𝜇𝑛 (x𝑖 (𝑡)))𝜇𝑛 (x𝑖 (𝑡))/𝑁 a vector b𝑗 (𝑡) to it characteristics. such that b𝑗 (𝑡)x(𝑡) The = 0. system Thus, impact onnormal the gait is well suited×for𝑖=1 the ∑𝑁 0&2&(2&0 2">6)9 , 23& "#$ 23& 3793&' 68 ?6,7>& 16,<&(2% 2'"(= 68 %7D#C I3& %"?& 6((G'% E723 (>G%2&' !F 23"2 7% 6((>G0&0 the following homogeneous polynomial of degree 𝑛 in )G?,&' 𝐷 + 𝛽 ∥𝜇 (x (𝑡))∥ /𝑁 𝑛 𝑖 𝑖=1 ical routine because there is no need to equip the observed person with Domain rated as the most important by patients. ;?. walking. ?. visual function. thinking and memory. Neuropsychological Magnetic Resonance parameter and patient Imaging (MRI) with related outcomes Gd contrast enhancer. Progression of disability. Relapses. bladder control. disability. lack of pain. disability. vKkvms† sKImsJKJ† Imkx[k}m}v`‚† EkJ† KEIZ† EIx[Ex[mk† mM† E† ) † vKkvms† €Ev† eEs^KJ† €[xZ[k† xZK† JExE† vxsKEe

(77) † '[X}sK† † vZm€v† E† [v}E`[ƒEx[mk† mM† /9;† [k† E† #EsxKv[Ek† ImmsJ[kExK† v‚vxKe† Ex† xZK† s[XZx

(78) † "`EI^† Jmxv† [kJ[IExK† eKEv}sKJ† Kk[smkeKkx

(79) † 𝑖 #[sI`Kv† 𝑦 `KX† EkJ† vr}EsKv† [kJ[IExK† vxEkJ† oZEvKv† Mms† xZK† `KMx† EkJ† 𝑥 s[XZx† 𝑖 vmeKmkK† 𝑖 𝑖 <ZK†𝑖Hm`J† sKImXk[ƒKJ† Msme† €E`^[kY† E`mkX† xZK† Smms

(80) † JEvZKJ† `[kK† [kJ[IExKv† xZK† [JKE`† €E`^[kX† oExZ† Mms† xZ[v† eKEv}sK„ eKkx† xm† €Z[IZ† 𝑁 J[OKsKkIKv† Mms† KEIZ† vxEkJ† oZEvK† EsK† Imeo}xKJ† 𝑖 𝑖 𝑖v†E†I`[k[IE``‚† [k†msJKs†xm†Imeo}xK†xZK†vxsE[XZxkKvv†mM†€E`^[kX

(81) † 𝑖=1 E`[JExKJ†sKMKsKkIK†E`}K†KEIZ†oEsx[I[oEkx†Imeo`KxKJ†v[†<[eKJ† =o† k (m† EvvKvveKkx† xKvxv† }v[kX† xZK† E<=(† EooEsEx}v† HKMmsK† EkJ† Mm}s† xKvxv† EWKs† xZK† RK`J† xs[E`† [k† Ek† }kmHvxs}IxKJ† smme

(82) † 2Ks† E† oKs[mJ† mM† RK† €KK^v† E† xmxE`† Eem}kx† mM† † ("† vKkvms† sKImsJ[kXv†€KsK†Im``KIxKJ

(83) †<ZKvK†JExE†[kI`}JKJ†mKsE``†† EIx[Ex[mkv†mM†) †vKkvmsv†Msme†€Z[IZ††xsEKsvKJ†€E`^[kX† oExZv† Im}`J† HK† JKxKIxKJ

(84) † † J[OKsKkx† €E`^[kX† oExZv† €KsK† JKRkKJ† [k† E``† SExv† xmXKxZKs

(85) † )m€KKs † Mms† xZK† o}somvK† mM† xZ[v† 𝐷 KE`}Ex[mk † mk`‚† JExE† sKImsJKJ† [k† xZK† M}``‚ Kr}[ooKJ† SEx† EsK† 𝑗 Imkv[JKsKJ

(86) † ,k† xZ[v† SEx † † €E`^[kX† oExZ† xsEkv[x[mkv† 𝐷 Im}`J† HK† eKEv}sKJ† H‚† ) † vKkvmsv† EkJ† /9;† Imkx[k}m}v`‚

(87) † "KMmsK† KE`}Ex[kX † sKImsJ[kXv† `mkXKs† xZEk† † vKImkJv† EkJ† vZmsxKs† xZEk† † vKImkJv† €KsK† KI`}JKJ† J}K† xm† o`E}v[H[`[x‚

(88) † 9KImsJ[kXv† m}xv[JK†xZmvK†HmsJKsv†Jm†[k†emvx†IEvKv†K[xZKs†sKqsKvKkx†km†I`KEs† emKeKkx† `[^K† K

(89) X

(90) † }v[kX† E† EI}}e† I`KEkKs† [k† xZK† Smms† ms†Jm† kmx†osm[JK†Kkm}XZ†JExE†Mms†EkE`‚v[v

(91) †<ZK†R`xKs[kX†`KEKv†† sKImsJ[kXv†Mms†KE`}Ex[mk

(92) †'ms†xZmvK†sKImsJ[kXv†xZK†J}sEx[mk†Mms† xZK† xsEkv[x[mkv† Ev† eKEv}sKJ† H‚† xZK† ) † vKkvmsv† EkJ† xZK† /9;† EkJ† xZK† sKv}`x[kX† XE[x† K`mI[x[Kv† EsK† Imeo}xKJ

(93) † =v[kX† mk`‚† xZK† JExE† sKImsJKJ† H‚† xZK† /9;† xZK† sKE`† J[vxEkIK† xsEKsvKJ† EkJ† xZK† r}E`[x‚† Is[xKs[E† vImsKv† Mms† €E`^[kX† xZKsKaHEI^† EkJ† Mms† x}sk[kX† EsK† Imeo}xKJ

(94) †. power and coordination of hands. swallowing. time. time. speech. sexuality. normal skin sensation. Modified MSFC:(not EDSS!) - ambulation T25FW,6MW - upper limbs (9HPT) - cognition PASAT-3,SDMT - vision (LCSLC). motor Fatigue (FSMC) cognitive Depression (HADS) Anxiety (HADS) [Quality of life (MSIS-29)]. ⇓. ⇓. ⇓. ⇓. bowel control mood. awakefulness and alertness. 4×. 0 10 20 30 40 0 10 20 30 40 percentage of MS patients (left: <5 years, right: >15 years). 07%5>"4&0 %&AG&)(& E"% 68 86G'F E37>& 23& '&"> )G?,&' E"% sensors or markers. It avoids the use of a treadmill, allows to analyse the gait both during straight lines and turns, and to analyze long walking tests, therefore allowing for an analysis of the motor fatigability [10].. =. no change in therapy reevaluate in 6 months. 1×. =. reevaluate before 3 months. 2×. or 1×. =. consider optimization or change of therapy. ,4 (>G%2&' % %&K&'"> (4(>&% !B79%C *!=# 26 *!>##C. 

(95) /. Figure 12: GAIMS measures feet trajectories with range laser scanners and derives many gait characteristics.. Figure 13: I-see-3D can project on the floor, in realtime, the feet trajectories observed with GAIMS.. Measuring the trajectories of the lower limbs extremities is sufficient, as we demonstrated that GAIMS measures an important quantity of information about the state of the patient, which is done with an adequate accuracy. In particular, GAIMS is able, with the help of machine learning techniques, to better detect ataxia than gait disorder specialists [11], and to detect when ataxia is increased between two successive visits of the same patient [12]. Moreover, it is possible to derive scores based on objective and quantitative measures of the gait that are well correlated with the subjective score used by neurologists [13]. Finally, GAIMS has been able to detect significant differences between healthy people, and different disability levels groups of people with MS [14].. References [1] S. Piérard, S. Azrour, and M. Van Droogenbroeck. Design of a reliable processing [6] M. Teixidó, T. Pallejà, M. Tresanchez, M. Norgués, and J. Palacín. Measuring pipeline for the non-intrusive measurement of feet trajectories with lasers. In IEEE oscillating walking paths with a LIDAR. Sensors, 11:5071–5086, 2011. International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages [7] S. Lejeune. Reconnaissance de personnes sur base des caractéristiques de la marche 4399–4403, Florence, Italy, May 2014. observées avec des capteurs laser. Master’s thesis, University of Liège, Belgium, 2014. [2] X. Song, X. Shao, H. Zhao, J. Cui, R. Shibasaki, and H. Zha. An online approach: [8] C. Heesen, J. Böhm, C. Reich, J. Kasper, M. Goebel, and S. Gold. Patient perception Learning-semantic-scene-by-tracking and tracking-by-learning-semantic-scene. In IEEE of bodily functions in multiple sclerosis: gait and visual function are the most valuable. International Conference on Computer Vision and Pattern Recognition (CVPR), pages Multiple Sclerosis, 14:988–991, 2008. 739–746, San Francisco, USA, June 2010. [9] M. Stangel, I. Penner, B. Kallmann, C. Lukas, and B. Kieseier. Towards the [3] M. Mucientes and W. Burgard. Multiple hypothesis tracking of clusters of people. In implementation of ’no evidence of disease activity’ in multiple sclerosis treatment: the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages multiple sclerosis decision model. Therapeutic Advances in Neurological Disorders, 692–697, Beijing, China, October 2006. 8(1):3–13, January 2015. [4] O. Barnich, S. Piérard, and M. Van Droogenbroeck. A virtual curtain for the detection [10] S. Piérard, R. Phan-Ba, and M. Van Droogenbroeck. Understanding how people with of humans and access control. In Advanced Concepts for Intelligent Vision Systems MS get tired while walking. Multiple Sclerosis Journal, 23(S11):406, September 2015. (ACIVS), Part II, pages 98–109, Sydney, Australia, December 2010. Proceedings of ECTRIMS 2015 (Barcelona, Spain), P812 (selected as a "top score [5] T. Frenken, M. Lipprandt, M. Brell, M. Gövercin, S. Wegel, E. Steinhagen-Thiessen, poster"). and A. Hein. Novel approach to unsupervised mobility assessment tests: Field trial for [11] S. Piérard, R. Phan-Ba, and M. Van Droogenbroeck. Machine learning techniques to aTUG. In 6th International Conference on Pervasive Computing Technologies for assess the performance of a gait analysis system. In European Symposium on Artificial Healthcare (PervasiveHealth), pages 131 –138, San Diego, USA, May 2012. Neural Networks, Computational Intelligence and Machine Learning (ESANN), pages 419–424, Bruges, Belgium, April 2014.. [12] S. Piérard, S. Azrour, R. Phan-Ba, and M. Van Droogenbroeck. Detection and characterization of gait modifications, for the longitudinal follow-up of patients with neurological diseases, based on the gait analyzing system GAIMS. In BIOMEDICA (the European Life Sciences Summit), Maastricht, The Netherlands, June 2014. [13] S. Azrour, S. Piérard, P. Geurts, and M. Van Droogenbroeck. Data normalization and supervised learning to assess the condition of patients with multiple sclerosis based on gait analysis. In European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN), pages 649–654, Bruges, Belgium, April 2014. [14] S. Piérard, S. Azrour, R. Phan-Ba, V. Delvaux, P. Maquet, and M. Van Droogenbroeck. Diagnosing multiple sclerosis with a gait measuring system, an analysis of the motor fatigue, and machine learning. Multiple Sclerosis Journal, 20(S1):171, September 2014. Proceedings of ACTRIMS/ECTRIMS 2014 (Boston, USA), P232.. Acknowledgements. We are grateful to the volunteers involved in the project GAIMS, to the Walloon region of Belgium (www.wallonie.be) for partly funding it, and to BEA (www.bea.be) for the sensors. Samir Azrour is supported by a research fellowship of the Belgian National Fund for Scientific Research (www.frs-fnrs.be).. Human Motion Analysis for Healthcare Applications — London, UK — May 19th 2016.

(96)

Références

Documents relatifs

In conclusion, we can highlight the fact that machine-learning in general, and deep learning in particular, are now particularly common to address perception tasks for

In other words, in order to improve the sustainability of public places and spaces, the Topozym project opted for a methodology combining research, training and

Each sequence is classified as normal or abnormal according to this similarity measure and a threshold selected empirically, similarly to [2].. Compared to [2], we obtain same

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

Indeed, in Escherichia coli, resistance to nonbiocidal antibio fi lm polysaccharides is rare and requires numerous mutations that signi fi cantly alter the surface

The word-fnal accent pattern, accompanied by the /+h/ afx and word-fnal aspiration, displays a shift in accent from the non-fnal to the fnal syllable1. When the /+h/ afx is

Les malades sont pris en charge par plusieurs centres : le service de pneumologie et des maladies infectieuses du centre hospitalier et universitaire Tokoin de Lomé et