research article a 2d markerless gait analysis methodology:...

Research ArticleA 2D Markerless Gait Analysis MethodologyValidation on Healthy Subjects

Andrea Castelli12 Gabriele Paolini12 Andrea Cereatti12 and Ugo Della Croce12

1Department of Information Engineering Political Sciences and Communication Sciences University of Sassari 07100 Sassari Italy2Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System Sassari Italy

Correspondence should be addressed to Andrea Cereatti acereattiunissit

Received 9 January 2015 Revised 5 April 2015 Accepted 7 April 2015

Academic Editor Reinoud Maex

Copyright copy 2015 Andrea Castelli et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

A 2D markerless technique is proposed to perform lower limb sagittal plane kinematic analysis using a single video camera Asubject-specific multisegmental model of the lower limb was calibrated with the subject in an upright standing position Anklesocks and underwear garments were used to track the feet and pelvis segments whereas shank and thigh segments were trackedby means of reference points identified on the model The method was validated against a marker based clinical gait modelThe accuracy of the spatiotemporal parameters estimation was found suitable for clinical use (errors between 1 and 3 of thecorresponding true values) Comparison analysis of the kinematics patterns obtainedwith the two systems revealed high correlationfor all the joints (082 lt 1198772 lt 099) Differences between the joint kinematics estimates ranged from 39 deg to 61 deg for thehip from 27 deg to 44 deg for the knee and from 30 deg to 47 deg for the ankle The proposed technique allows a quantitativeassessment of the lower limb motion in the sagittal plane simplifying the experimental setup and reducing the cost with respect totraditional marker based gait analysis protocols

1 Introduction

Three-dimensional (3D) marker-based clinical gait analysisis generally recognized to play an important role in theassessment therapy planning and evaluation of gait relateddisorders [1] It is performed by attaching physical markerson the skin of the subject and recording their position viamultiple cameras To date optoelectronic stereophotogram-metric systems represent the most accurate technology forthe assessment of joint kinematics [2] While the 3D char-acterization of motion represents the standard for clinicalgait analysis laboratories and research environments 3Dgait analysis remains underused in ambulatory environmentsdue to the costs time and technical requirements of thistechnology

Multicamera video-based markerless (ML) systems canrepresent a promising alternative to 3Dmarker-based systems[3ndash5] In fact the use of ML techniques does not require theapplication of fixtures on the skin of the patients [1] makingthe experimental sessions faster and simpler (eg do not

have to worry about markers falling off during the sessions)[1] 3D ML motion capture techniques have been extensivelypresented in [5ndash9] for different types of applications includ-ing clinical gait analysis and biomechanics However 3DML approaches similarly to 3D marker-based techniquesrequire the use of multiple cameras [5] specific calibrationprocedures time synchronization between cameras anda considerable dedicated space Furthermore the numberof cameras and their high image resolution dramaticallyincrease the computing time and resources required

When a 3D analysis is not strictly required a simpler 2Danalysis on the sagittal plane could be successfully used toquantify gait and to address specific clinical questions [10 11]A single-camera approach is sufficient for the description ofgait in 2D and allows for a simplified experimental setupreducing the space needed for the equipment the numberof cameras and the costs associated Video recording incombination with observational gait evaluation scales [11] iscommonly employed to perform visual and qualitative gaitanalysis In this regard amethodology to quantitatively assess

Hindawi Publishing CorporationComputational and Mathematical Methods in MedicineVolume 2015 Article ID 186780 11 pageshttpdxdoiorg1011552015186780

2 Computational and Mathematical Methods in Medicine

joint kinematics from the video footage can provide addedvalue to the patient care with no extra resources involved

Single-camera ML techniques have been mainly devel-oped for motion recognition and classification [12ndash19] Tothe best of our knowledge only a paucity of studies haveproposed ML methods for the estimation of the lower limbjoint kinematics for clinical evaluations [20ndash24] However inthe abovementioned studies the analyses were either limitedto a single joint [20 21] or they lacked a validation againsta clinically accepted gold standard [22 23] These limitationshampered the wide spread use of such techniques in clinicalsettings [4 7 9]

In this study we present a novel 2D model-based MLmethod for clinical gait analysis using a single camera theSEGMARK method It provides unilateral joint kinematicsof hip knee and ankle in the sagittal plane along withthe estimation of gait events and spatiotemporal parametersThe method uses garments (ie socks and underwear) wornby the subjects as segmental markers to track the pelvisand feet segments The segmentsrsquo model templates and therelevant anatomical coordinate systems are calibrated in astatic reference image from anatomical landmarks manuallyidentified by an operatorThemethod applicability was testedand the performance evaluation was carried out on tenhealthy subjects walking at three different speeds using anoptoelectronic marker-based system as gold standard

2 Materials and Methods

21 Experimental Protocol and Setup Ten healthy subjects(males age 33plusmn3 yo) wearing only homogeneously coloured(white) and adherent ankle socks and underwear were askedto walk at comfortable slow and fast speed along a straight 8-meter walkway An RGB video camera (Vicon Bonita Video720c 1280 times 720 p 50 fps) was positioned laterally to thewalkway A homogenous blue background was placed oppo-site to the camera To prevent blurred images the exposuretime was set to a low value (5ms) and the illumination levelwas set accordinglyThe image coordinate system (CSI) of thevideo camera was aligned to the sagittal plane identified bythe direction of progression and the vertical direction

Three trials per subject were captured for each gait speedThe starting line was set so that the foot in the foregroundcould enter the field of view first and hit the ground whenfully visible A static reference image with the subject inan upright standing position centered in the field of viewof the camera was captured prior to each experimentalsession The subjects were then asked to walk along a linedrawn on the floor placed at a known distance from theimage plane identical to the distance between the cameraand the subject during the static reference image acquisitionFor validation purposes 3Dmarker-based data synchronouswith the video data was captured at 50 fps using a 6-camerastereophotogrammetric system (Vicon T20) Retroreflectivespherical markers (14mm diameter) were attached to thesubjects according to the Davis model [26] provided by theVicon Nexus software (Plug in Gait)

22 Image Preprocessing Camera lens distortion was cor-rected using the Heikkila undistortion algorithm [27] Thespatial mapping of the camera image was determined byassociating the measured foot length to the foot segmentalmarker length expressed in pixels from the static referenceimage (1 pixel asymp 1mm)

To separate the moving subject from the background asegmentation procedure based on background subtraction inthe HSV color space was applied [6] The underwear andankle socks were extracted using a white color filter andused as segmental markers An automatic labelling processto identify the segmental markers was performed The pelvissegmental marker was identified as the group of white pixelswith higher vertical coordinates in theCSIThe feet segmentalmarkers were identified and tracked using the predictedpositions of their centroids based on their velocity at theprevious two frames Cannyrsquos edge operator [28] was usedto obtain the silhouette (Figure 1) and the segmental markerscontours

23 Cycle Segmentation and Gait Parameters DeterminationHeel strike and toe off events were automatically estimatedusing a method originally developed for marker-based sys-tems [29] and adapted to our ML method Per each timeframe the centroids of the pelvis and both feet were deter-mined The gait events (heel strike and toe off) instants weredetermined when the maximum horizontal relative distancebetween the pelvis and the foot centroids is achieved

An expert operator manually identified the same gaitevents using the video footage and the heel and toe 3Dmarker trajectories as reference The following spatial andtemporal parameters were then calculated for both ML andmarker-based data cadence walking speed stride time andstride length The estimated parameters were compared forvalidation purposes

24 Model Calibration A subject-specific multisegmentalmodel of the lower limb was used to track the segments andto compute the relevant joint kinematics The model is madeof four segments (foot tibia femur and pelvis) connected byhinges The position of the following anatomical landmarkswasmanually identified by the operator in the static referenceimage lateral malleolus (LM) lateral femoral epicondyle(LE) and greater trochanter (GT) (Figure 2(a))

The foot model template was defined as the posteriorhalf of the foot segmental marker contour The anatomicalcoordinate system (CSA) of the foot was defined on thefoot model template with the positive 119909-axis coincidentwith the line fitting the lower-posterior contour and orientedtowards the toes The origin was made to coincide with themost posterior point of the foot segmental marker contourA technical coordinate system (CST) was defined with the119909-axis coinciding with the corresponding axis of the CSIand centred in LM (Figure 2(b)) The transformation matrixbetween the CST andCSA of the foot was computed fCSATfCST

The CSA of the tibia was defined with the 119910-axis joiningLM with LE (origin on LM) The tibia model templatewas defined based on ten reference points identified on the

Computational and Mathematical Methods in Medicine 3

(a) (b) (c)

Figure 1 Contour extraction and silhouette contour deformation for three different gait cycle percentages It can be noticed that in cases (b)and (c) there is an overlap between the foreground and background legs that prevents the identification of the correspondent boundaries ofthe segments

PL

GT

LE

LMrshk

rtik

(a)

LMCST

CSA

(b)

PLfirstPPfirst

PLlastPPlast

(c)

Figure 2 (a) Anatomical landmarks and anatomical coordinate systems (CSA) for the segments analyzed femur CSA (yellow axes) and femurreference points (yellow points) identified by the yellow arcs tibia CSA (cyan axes) and tibia reference points (cyan points) identified by thecyan arcs (b) Foot CSA and model template (c) Double anatomical calibration of the most lateral point (PL) in the first and last frame of thegait cycle

silhouette as the intersections between the shank contour andthe circles of radius 119903sh119896 centered in LM (Figure 2(a)) Thelength of the imposed radii was chosen so that the referencepoints would fall within the middle portion of the shanksegment (between the 25 and 75 of the segment length)This avoided the reference points to fall on the portions ofthe segment adjacent to the joints These areas are in factsubject to a larger soft tissue deformation during gait [2]Thetibia CST was defined with the 119909-axis parallel to the 119909-axisof the CSI and centred in the centroid of the tibia referencepoints The transformation matrix between CST and CSA ofthe tibia was computed tibCSATtibCST (Figure 2(a)) LM and LEpositions and the tibia reference points were then expressedin the tibia CST (tibCSTp0

119896 119896 = 1 10)

The CSA of the femur was defined with the 119910-axis joiningLE with GT (origin on LE) The femur model templatewas defined based on six reference points identified on thesilhouette as the intersections between the thigh contour andthe circles of radius 119903th119896 centered in LE (Figure 2(a)) Thefemur CST was defined with the 119909-axis parallel to the 119909-axisof the CSI and centred in the centroid of the thigh referencepoints The transformation matrix between the CST and CSAof the femur was computed femCSATfemCST (Figure 2(a)) LEand GT positions and the femur reference points were thenexpressed in the femur CST (femCSTp0

119896 119896 = 1 6)

The pelvis CSA origin was set in the most lateralpoint (PL) of the pelvis segmental marker upper contour(Figure 2(a)) The 119909-axis was oriented as the line fitting


the portion of the pelvis segmental marker upper contourdefined from PL plusmn 20 pixels and pointing towards thedirection of progression Due to both parallax effect andpelvis axial rotation during gait the shape of the pelvissegmental marker changes remarkably throughout the gaitcycle (Figure 2(c)) To improve the quality of the PL trackingprocess a double calibration approach was implemented Atthe first and last frames of the gait cycle the PL positions weremanually identified by the operator (PLfirst PLlast) whereasthe positions of the most posterior point (PPfirst and PPlast)of the pelvis upper contour were automatically identifiedThe horizontal distances between PLfirst and PPfirst (119889119909first)and between PLlast and PPlast (119889119909last) were calculated andthe incremental frame by frame variation Δ was computedaccording to

Δ =

1003816100381610038161003816119889119909last minus 119889119909first1003816100381610038161003816

119872 (1)

with119872 representing the number of framesThe incremental frame variation was then used to com-

pute the distance 119889119909119894

between the points PP and PL incorrespondence of the 119894th frame

119889119909119894= 119889119909first + (Δ sdot 119894) (2)

The position of PL at the 119894th instant (PL119894) was determined

from the automatically detected position of PP at the sametime instant (PP

119894)

PL119894= PP119894+ 119889119909119894 (3)

25 Dynamic Processing The dynamic trials were processedusing a bottom-up tracking approach starting from the footand moving up the chain The foot was tracked using aniterative contour subtraction matching technique betweenthe contour at the 119894th frame and the template foot contourAt the 119894th frame the foot CST was rotated and translatedaround the prediction of the LMposition based on its velocityestimated in the previous two frames To make the processmore efficient the rotational span limits were defined basedon the gait cycle phase plusmn10 degrees during the stance phaseand plusmn30 degrees during the swing phase The transformationmatrix fCSTT(119894)fCSI between the foot CST and the CSI wasdetermined by maximizing the superimposition between thefoot template and the foot contour at the 119894th frame (ie theminimum number of active pixels resulting from the imagesubtraction) The transformation matrix between the footCSA and the CSI was computed as

fCSAT (119894)fCSI =fCSAT (119894)fCST sdot

fCSTT (119894)fCSI (4)

Due to leg superimposition and soft tissues deformationthe silhouette contours of the shank and thigh change duringthe gait cycle making the tracking difficult (Figure 1) Toovercome this issue the following procedure was adopted forthe tibia At the 119894th frame a registration of first approximationbetween the tibia CST and the CSI was carried out using theposition of LM and the prediction of LE An approximatedestimate of the tibia reference points positions with respect

Contourarc

LM

119858k1198490k

Figure 3 Tibia reference points detection These points are identi-fied on the silhouette as the intersections between the shank contourand the circles of radius 119903sh119896 centered in LM LM (cyan circle)predicted LE (white cross)magnified arc of circumference of radius119903sh119896 (yellow curve) y

119896 reference point detected in the current frame

(red circle) p0119896 template reference point after fitting (cyan circle)

silhouette contour line (contour)

to the CSI was then obtained and a 10 times 10 pixels region ofinterest was created around each point The final referenceposition vectors CSIy

119896of the tibia were detected as the

intersection where available between the portion of theshank contour included in the region of interest and thecircle of radius 119903tib119896The transformationmatrix tibCSTT(119894)tibCSIbetween the CSI and CST was determined using a SingularValue Decomposition procedure [25] between the positionvectors tibCSTp0

119896and the corresponding points tibCSIy

0 Due to

leg superimposition the number of points tibCSTp0119896involved

in the fitting procedure varied according to the number ofintersections tibCSIy

0available (Figure 3) The transformation

matrix between the tibia CSA and the CSI was computed as

tibCSAT (119894)tibCSI =tibCSAT (119894)tibCST sdot

tibCSTT (119894)tibCSI (5)

An identical procedure was employed to determine thetransformation between the CSI and CSA of the femurfemCSAT(119894)femCSI at the 119894th frame

From the relevant transformation matrices the jointangles between the pelvis and the femur (hip flexext)between the femur and the tibia (knee joint flexext) andbetween the tibia and the foot (ankle joint flexext) werecomputed

26 Data Analysis For each gait speed the accuracy ofthe spatiotemporal gait parameters estimated by the MLapproach was assessed in terms of the mean absolute error(MAE) and MAE over trials and subjects (3 times 10 trials)Both the ML and marker-based angular kinematic curveswere filtered using a fourth-order Butterworth filter (cut-offfrequency at 10Hz)The sagittal angular kinematics producedby the Plug in Gait protocol was used as gold standard [26]

The kinematic variables were time-normalized to thegait cycle Furthermore for each gait trial and each joint


Table 1 Gait spatiotemporal parametersMean absolute (MAE) andpercentage errors of the cadence walking speed stride time andstride length for different gait speeds (slow comfortable and fast)

Slow speed Comfortable speed Fast speedMAE MAE MAE MAE MAE MAE

Cadence(stepsmin) 093 1 134 1 223 2

Walkingspeed (ms) 002 2 003 3 005 3

Stride time(s) 001 1 001 1 002 2

Stridelength (m) 003 3 004 3 005 3

the average root-mean-square deviation (RMSD) valuebetween the joint kinematic curves estimated by the MLmethod and the gold standard were computed over the gaitcycle and averaged across trials and subjects The similaritybetween the curves provided by the MLmethod and the goldstandard was assessed using the linear-fit method proposedby Iosa and colleagues [30] The kinematic curves producedby the ML system were plotted versus those produced by thegold standard and the coefficients of the linear interpolationwere used to compare the curves in terms of shape similarity(1198772) amplitude (angular coefficient1198600) and offset (constantterm1198601)The average of1198600 and1198601 across trials and subjectswas calculated for each gait speed

3 Results

The results relative to the spatiotemporal gait parameters areshown in Table 1 For all parameters and gait speeds theerrors were between 1 and 3 of the corresponding truevalues

Results relative to the joint kinematics are shown inTable 2 Average RMSD values over trials and subjectsranged from 39 deg to 61 deg for the hip from 27 deg to44 deg for the knee from 30 deg to 47 deg for the ankle andfrom 28 deg to 38 deg for the pelvic tilt

The results of the linear-fit method (1198772) highlightedexcellent correlation (from 096 to 099 for all gait speeds) forhip and knee joint kinematics The ankle kinematics showedstrong correlation (from 082 to 087 for all gait speed)Conversely the pelvic tilt showed no correlation The hipand ankle joint kinematics were underestimated in terms ofamplitude (1198600 ranged from 073 to 076 and from 082 to 087for the hip and ankle resp) whereas the knee kinematicswere overestimated (1198600 from 108 to 111) The values of theoffset 1198601 were consistent amongst the different gait speeds(maximum difference of 24 deg for the hip joint across allgait velocities) and ranged from minus01 deg to 23 deg fromminus77 deg tominus64 deg and fromminus47 deg tominus33 deg for the hipknee and ankle respectively The joint kinematics and pelvictilt curves averaged over trials and subjects and normalisedto the gait cycle are reported in Figures 4 5 6 and 7

4 Discussion and Conclusion

The aim of this study is to implement and validate a MLmethod based on the use of a single RGB camera forlower limb clinical gait analysis (SEGMARK method) Theestimated quantities consist of hip knee and ankle jointkinematics in the sagittal plane pelvic tilt and spatiotemporalparameters of the foreground limb The SEGMARK methodis based on the extraction of the lower limbs silhouette andthe use of garment segmental markers for the tracking of thefoot and pelvis

Key factors of the proposed method are the use ofanatomically based coordinate systems for the joint kinemat-ics description the automatic management of the superim-position between the foreground and background legs and adouble calibration procedure for the pelvis tracking

For a clinically meaningful description of the joint kine-matics it is required for each bone segment to define acoordinate system based on the repeatable identification ofanatomical landmarks [31] In our method the anatomicalcalibration is performed by an operator on the static referenceimage by clicking on the relevant image pixels Althoughit might sound as a limitation the manual anatomicallandmarks calibration is a desired feature because it allowsthe operator to have full control on the model definition

When using a single RGB camera for recording gaitthe contours of the foreground and background legs areprojected onto the image plane and their superimpositionmakes difficult the tracking of the femur and tibia segmentsduring specific gait cycle phases (Figure 1) To overcomethis problem the model templates of the femur and tibiasegments werematched for each frame to an adaptable set oftarget points automatically selected from the thigh and shankcontours

The use of a silhouette-based approach implies that noinformation related to the pixel greyscale intensity or colorvalues are exploited except for the background subtractionprocedure [7 8] This should make the method less sensitiveto change in light conditions or cameras specifications [7 8]

The accuracy with which the spatiotemporal parametersare estimated (Table 1) is slightly better than other MLmethods [14 15] Furthermore the percentage error wasalways lower than 3 The errors associated to the gaittemporal parameter (stride time) estimation were less than002 s for all gait speeds

The kinematic quantities estimated by the SEGMARKmethod and those obtained using a clinically validated gaitanalysis protocol were compared in terms of root-mean-square deviation (RMSD) shape (1198772) amplitude (1198600) andoffset (1198601) (Table 2) In this respect it is worth noting thatthe differences found are not entirely due to errors in the jointkinematics estimates but also to the different definitions usedfor the CSA and the different angular conventions adopted(2D angles versus 3D Euler angles)

Overall the RMSD values ranged between 27 and 61 degacross joints and gait speedsThe smallest RMSD values werefound for the knee joint kinematics followed by the hipand ankle joints kinematics We reckon that the differences


Table 2 Lower limb joint and pelvis kinematicsThe average root-mean-square deviation (RMSD) value between the joint kinematics curvesestimated by the ML method and the gold standard are computed over the gait cycle and averaged across trials and subjects The similaritybetween the curves obtained with the proposedMLmethod and the gold standard is assessed using the linear-fitmethod [25]The coefficientsof the linear interpolation were used to compare the curves in terms of shape similarity (1198772) amplitude (angular coefficient 1198600) and offset(constant term 1198601) The average of 1198600 and 1198601 across trials and subjects is calculated for each gait speed

SpeedHip Knee Ankle Pelvis

RMSD(deg) 119877

2 11986001198601

(deg)RMSD(deg) 119877

2 11986001198601


2 11986001198601


2 11986001198601

(deg)Slow 39 97 76 102 27 99 108 minus636 32 84 82 minus467 28 06 05 minus304Comfortable 48 97 76 minus09 36 99 111 minus771 30 87 74 minus364 30 01 38 minus671Fast 61 96 73 231 44 98 110 minus647 47 82 68 minus335 38 18 minus72 minus19

Slow

Fast

Comfortable

30

20

10

0

0

00

Exte

nsio

nFl

exio

n

10050

0 100500 10050

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

(deg

)

Exte

nsio

nFl

exio

n(d

eg)

Exte

nsio

nFl

exio

n(d

eg)

Gait cycle () Gait cycle ()

Gait cycle ()

Figure 4 Hip flexionextension averaged over subjects and trials for the selected gait speed (average solid lines SD shaded area red = MLblue = Plug in Gait protocol)


Slow

Fast

Comfortable

Flex

ion

0 10050

(deg

)

Gait cycle ()

0 10050

Gait cycle ()0 10050

Gait cycle ()

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

Flex

ion

(deg

)

Flex

ion

(deg

)

Exte

nsio

n

Exte

nsio

n

Exte

nsio

n

Figure 5 Knee flexionextension averaged over subjects and trials for the selected gait speed (average solid lines SD shaded area red =MLblue = Plug in Gait protocol)

at the hip angle between the SEGMARK method and thePlug in Gait model were mainly due to inaccuracies in thepelvis tracking which proved to be a critical aspect of thesingle-camera technique proposed This is mainly due tointrinsic difficulties associated to the pelvis tracking suchas the small range of motion in the sagittal plane duringgait (lt5 deg) the wider range of motion in the frontal andhorizontal planes and the difficulty to isolate the pelvis fromthe lower trunkThe differences in the ankle joint kinematics

observed between the SEGMARK method and the Plug inGait model (Table 2) can be probably ascribed to the 2Dtracking of the foot segment In fact the SEGMARKmethodcannot account for the foot motion in the transverse planeleading to an underestimation of the ankle joint motionamplitude (1198600 varied from 082 to 068 for increasing gaitspeed)The waveform similarity analysis highlighted that theamplitude of the knee joint angle was consistently overes-timated (1198600 equals to 111 for comfortable speed) whereas


Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5

Figure 6 Ankle plantardorsiflexion averaged over subjects and trials for the selected gait speed (average solid lines SD shaded area red =ML blue = Plug in Gait protocol)

the hip and ankle joint kinematic curves were consistentlyunderestimated (1198600 equals to 076 and 074 for hip and ankleresp for comfortable speed) This can be explained by theconsistent overestimation of the pelvis motion which was inphase with the femur angular displacement combined withthe underestimation of the foot motion in the sagittal planeAn increase of the gait speed showed a negative effect on theaccuracy of the kinematics estimation for all the joints dueto the significantly smaller number of frames recorded at the

fast gait (asymp40 frames) compared to the slow gait (asymp80 frames)This result underlines the importance of a sufficiently highframe rate when recording the video footage

To our knowledge amongst the 2D ML methods forclinical gait analysis proposed in the literature just a few useanatomically relevant points to define the CSA [22 32 33]Goffredo and colleagues [22] used an approach based onskeletonisation and obtained the proportions of the humanbody segments from anatomical studies [34] This approach


Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2

Figure 7 Pelvic tilt averaged over subjects and trials for the selected gait speed (average solid lines SD shaded area red = ML blue = Plugin Gait protocol)

although completely automatic neglects the subject-specificcharacteristics possibly leading to joint angles estimationinaccuracies Other recent ML validation studies rely uponthe built-in algorithmof theMicrosoftKinect to identify jointcentres [33 35] which are therefore not based on anatomicalinformation making the use of such techniques questionablefor clinical applications Another important limitation ofthe 2D ML methods previously proposed is the lack of asystematic validation via a well-established gold standard for

clinical gait analysis applications [13ndash19] To our knowledgeonly the work presented by Majernik [32] compared thejoint kinematics with the sagittal plane joint angles producedby a simultaneously captured 3D marker-based protocolHowever neither reference to the specific clinical gait analysisprotocol used nor a quantitative assessment of the methodperformance is reported Similarly Goffredo and colleagues[22] only performed a qualitative validation of their methodby comparing the estimated joint kinematics to standard


patterns taken from the literature Moreover none of theabovementionedworks described the procedure used to trackthe pelvis which is critical for the correct estimation of thehip joint angle

Interestingly when comparing our results with the jointsagittal kinematics obtained from more complex 3D MLtechniques we found errors either comparable or smallerSpecifically Sandau et al [3] reported a RMSD of 26 deg forthe hip 35 deg for the knee and 25 deg for the ankle whichare comparable with the RMSD values reported in Table 2 forthe comfortable gait speed Ceseracciu et al [4] reported aRMSD of 176 deg for the hip 118 deg for the knee and 72 degfor the ankle sensibly higher than the values reported in thiswork This further confirms the potential of the SEGMARKapproach for the study of the lower limbmotion in the sagittalplane for clinical applications

This study has limitations some of them inherent tothe proposed ML technique whereas others related to theexperimental design of the study First the joint kinematicsis only available for the limb in the foreground whileother authors managed to obtain a bilateral joint kinematicsdescription using a single camera [22] Therefore to obtainkinematics from both sides the subject has to walk inboth directions of progression Second the segmentationprocedure for the subject silhouette extraction takes advan-tage of a homogeneous blue background This allows foroptimal segmentation results but it adds a constraint in theexperimental setup For those applications where the useof a uniform background is not acceptable more advancedsegmentation techniques can be employed [7ndash9] Finally theanatomical calibration procedure requires the operator tovisually identify the anatomical landmarks in the static imageand this operation necessarily implies some repeatabilityerrors which would need to be assessed in terms of inter andintra observer and inter and intrasubject variability [35] Asa future step the present methodology will be applied andvalidated on pathological populations such as cerebral palsychildren

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

References

[1] R Baker ldquoGait analysis methods in rehabilitationrdquo Journal ofNeuroEngineering and Rehabilitation vol 3 article 4 2006

[2] U Della Croce A Leardini L Chiari and A CappozzoldquoHuman movement analysis using stereophotogrammetry part4 assessment of anatomical landmark misplacement and itseffects on joint kinematicsrdquo Gait amp Posture vol 21 no 2 pp226ndash237 2005

[3] M Sandau H Koblauch T B Moeslund H Aanaeligs T Alkjaeligrand E B Simonsen ldquoMarkerless motion capture can providereliable 3D gait kinematics in the sagittal and frontal planerdquoMedical Engineeringamp Physics vol 36 no 9 pp 1168ndash1175 2014

[4] E Ceseracciu Z Sawacha and C Cobelli ldquoComparison ofmarkerless and marker-based motion capture technologiesthrough simultaneous data collection during gait proof ofconceptrdquo PLoS ONE vol 9 no 3 Article ID e87640 2014

[5] S Corazza L Mundermann E Gambaretto G Ferrigno andT P Andriacchi ldquoMarkerless motion capture through visualhull articulated ICP and subject specific model generationrdquoInternational Journal of Computer Vision vol 87 no 1-2 pp156ndash169 2010

[6] L Mundermann S Corazza A M Chaudhari T P Andri-acchi A Sundaresan and R Chellappa ldquoMeasuring humanmovement for biomechanical applications using markerlessmotion capturerdquo in 7th Three-Dimensional Image Capture andApplications vol 6056 of Proceedings of SPIE p 60560RInternational Society for Optics and Photonics San Jose CalifUSA January 2006

[7] L Mundermann S Corazza and T P Andriacchi ldquoThe evolu-tion of methods for the capture of human movement leadingto markerless motion capture for biomechanical applicationsrdquoJournal of NeuroEngineering and Rehabilitation vol 3 article 62006

[8] T B Moeslund and E Granum ldquoA survey of computer vision-based human motion capturerdquo Computer Vision and ImageUnderstanding vol 81 no 3 pp 231ndash268 2001

[9] H Zhou and H Hu ldquoHuman motion tracking forrehabilitationmdasha surveyrdquo Biomedical Signal Processing andControl vol 3 no 1 pp 1ndash18 2008

[10] A Harvey and J W Gorter ldquoVideo gait analysis for ambulatorychildren with cerebral palsy why when where and howrdquo Gaitamp Posture vol 33 no 3 pp 501ndash503 2011

[11] U C Ugbolue E Papi K T Kaliarntas et al ldquoThe evaluationof an inexpensive 2D video based gait assessment system forclinical userdquo Gait amp Posture vol 38 no 3 pp 483ndash489 2013

[12] J-H Yoo and M S Nixon ldquoAutomated markerless analysis ofhuman gait motion for recognition and classificationrdquo ETRIJournal vol 33 no 2 pp 259ndash266 2011

[13] R A Clark K J Bower B F Mentiplay K Paterson and Y-H Pua ldquoConcurrent validity of the Microsoft Kinect for assess-ment of spatiotemporal gait variablesrdquo Journal of Biomechanicsvol 46 no 15 pp 2722ndash2725 2013

[14] N Vishnoi Z Duric and N L Gerber ldquoMarkerless identifica-tion of key events in gait cycle using image flowrdquo in Proceedingsof the Annual International Conference of the IEEE Engineeringin Medicine and Biology Society (EMBC rsquo12) pp 4839ndash48422012

[15] N R Howe M E Leventon and W T Freeman ldquoBayesianreconstruction of 3Dhumanmotion from single-camera videordquoin Advances in Neural Information Processing Systems pp 820ndash826 MIT Press 1999

[16] D KWagg andM S Nixon ldquoAutomated markerless extractionof walking people using deformable contourmodelsrdquoComputerAnimation and Virtual Worlds vol 15 no 3-4 pp 399ndash4062004

[17] C Bregler and M Jitendra ldquoTracking people with twists andexponential mapsrdquo in Proceedings of the IEEE InternationalConference on Computer Vision and Pattern Recognition 1998pp 8ndash15 Santa Barbara Calif USA June 1998

[18] M Enzweiler and D M Gavrila ldquoMonocular pedestrian detec-tion survey and experimentsrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 31 no 12 pp 2179ndash21952009


[19] J E Boyd and J J Little ldquoBiometric gait recognitionrdquo inAdvanced Studies in Biometrics vol 3161 of Lecture Notes inComputer Science pp 19ndash42 Springer Berlin Germany 2005

[20] E Surer A Cereatti E Grosso and U D Croce ldquoA markerlessestimation of the ankle-foot complex 2D kinematics duringstancerdquo Gait amp Posture vol 33 no 4 pp 532ndash537 2011

[21] A Schmitz M Ye R Shapiro R Yang and B NoehrenldquoAccuracy and repeatability of joint angles measured usinga single camera markerless motion capture systemrdquo Gait ampPosture vol 47 pp 587ndash591 2014

[22] M Goffredo J N Carter and M S Nixon ldquo2D markerless gaitanalysisrdquo in Proceedings of the 4th European Conference of theInternational Federation for Medical and Biological Engineeringvol 22 of IFMBE Proceedings pp 67ndash71 Springer BerlinGermany 2009

[23] J Saboune and F Charpillet ldquoMarkerless human motion cap-ture for gait analysisrdquo httparxivorgabscs0510063

[24] A Leu D Ristic-Durrant and A Graser ldquoA robust markerlessvision-based human gait analysis systemrdquo in Proceedings of the6th IEEE International Symposium on Applied ComputationalIntelligence and Informatics (SACI rsquo11) pp 415ndash420 TimisoaraRomania May 2011

[25] F E VeldpausH JWoltring and L JMGDortmans ldquoA least-squares algorithm for the equiform transformation from spatialmarker co-ordinatesrdquo Journal of Biomechanics vol 21 no 1 pp45ndash54 1988

[26] R B Davis III S Ounpuu D Tyburski and J R Gage ldquoAgait analysis data collection and reduction techniquerdquo HumanMovement Science vol 10 no 5 pp 575ndash587 1991

[27] J Heikkila and O Silven ldquoCalibration procedure for short focallength off-the-shelf CCD camerasrdquo in Proceedings of the 13thInternational Conference on Pattern Recognition (ICPR rsquo96) vol1 pp 166ndash170 August 1996

[28] J F Canny ldquoA computational approach to edge detectionrdquo IEEETransactions on Pattern Analysis and Machine Intelligence vol8 no 6 pp 679ndash698 1986

[29] J A Zeni Jr J G Richards and J S Higginson ldquoTwo simplemethods for determining gait events during treadmill andoverground walking using kinematic datardquo Gait amp Posture vol27 no 4 pp 710ndash714 2008

[30] M Iosa A Cereatti A Merlo I Campanini S Paolucci and ACappozzo ldquoAssessment of waveform similarity in clinical gaitdata the linear fit methodrdquo BioMed Research International vol2014 Article ID 214156 7 pages 2014

[31] A Cappozzo U Della Croce A Leardini and L ChiarildquoHumanmovement analysis using stereophotogrammetry Part1 theoretical backgroundrdquoGaitampPosture vol 21 no 2 pp 186ndash196 2005

[32] J Majernik ldquoMarker-free method for reconstruction of humanmotion trajectoriesrdquo Advances in Optoelectronic Materials vol1 no 1 2013

[33] M Gabel R Gilad-Bachrach E Renshaw and A SchusterldquoFull body gait analysis with Kinectrdquo in Proceedings of the 34thAnnual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBS rsquo12) pp 1964ndash1967 SanDiego Calif USA September 2012

[34] R A Clark Y-H Pua A L Bryant andMAHunt ldquoValidity ofthe Microsoft Kinect for providing lateral trunk lean feedbackduring gait retrainingrdquo Gait amp Posture vol 38 no 4 pp 1064ndash1066 2013

[35] MH Schwartz J P Trost andRAWervey ldquoMeasurement andmanagement of errors in quantitative gait datardquoGait amp Posturevol 20 no 2 pp 196ndash203 2004

Submit your manuscripts athttpwwwhindawicom

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinsonrsquos Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom


joint kinematics from the video footage can provide addedvalue to the patient care with no extra resources involved

Single-camera ML techniques have been mainly devel-oped for motion recognition and classification [12ndash19] Tothe best of our knowledge only a paucity of studies haveproposed ML methods for the estimation of the lower limbjoint kinematics for clinical evaluations [20ndash24] However inthe abovementioned studies the analyses were either limitedto a single joint [20 21] or they lacked a validation againsta clinically accepted gold standard [22 23] These limitationshampered the wide spread use of such techniques in clinicalsettings [4 7 9]

In this study we present a novel 2D model-based MLmethod for clinical gait analysis using a single camera theSEGMARK method It provides unilateral joint kinematicsof hip knee and ankle in the sagittal plane along withthe estimation of gait events and spatiotemporal parametersThe method uses garments (ie socks and underwear) wornby the subjects as segmental markers to track the pelvisand feet segments The segmentsrsquo model templates and therelevant anatomical coordinate systems are calibrated in astatic reference image from anatomical landmarks manuallyidentified by an operatorThemethod applicability was testedand the performance evaluation was carried out on tenhealthy subjects walking at three different speeds using anoptoelectronic marker-based system as gold standard

2 Materials and Methods

21 Experimental Protocol and Setup Ten healthy subjects(males age 33plusmn3 yo) wearing only homogeneously coloured(white) and adherent ankle socks and underwear were askedto walk at comfortable slow and fast speed along a straight 8-meter walkway An RGB video camera (Vicon Bonita Video720c 1280 times 720 p 50 fps) was positioned laterally to thewalkway A homogenous blue background was placed oppo-site to the camera To prevent blurred images the exposuretime was set to a low value (5ms) and the illumination levelwas set accordinglyThe image coordinate system (CSI) of thevideo camera was aligned to the sagittal plane identified bythe direction of progression and the vertical direction

Three trials per subject were captured for each gait speedThe starting line was set so that the foot in the foregroundcould enter the field of view first and hit the ground whenfully visible A static reference image with the subject inan upright standing position centered in the field of viewof the camera was captured prior to each experimentalsession The subjects were then asked to walk along a linedrawn on the floor placed at a known distance from theimage plane identical to the distance between the cameraand the subject during the static reference image acquisitionFor validation purposes 3Dmarker-based data synchronouswith the video data was captured at 50 fps using a 6-camerastereophotogrammetric system (Vicon T20) Retroreflectivespherical markers (14mm diameter) were attached to thesubjects according to the Davis model [26] provided by theVicon Nexus software (Plug in Gait)

22 Image Preprocessing Camera lens distortion was cor-rected using the Heikkila undistortion algorithm [27] Thespatial mapping of the camera image was determined byassociating the measured foot length to the foot segmentalmarker length expressed in pixels from the static referenceimage (1 pixel asymp 1mm)

To separate the moving subject from the background asegmentation procedure based on background subtraction inthe HSV color space was applied [6] The underwear andankle socks were extracted using a white color filter andused as segmental markers An automatic labelling processto identify the segmental markers was performed The pelvissegmental marker was identified as the group of white pixelswith higher vertical coordinates in theCSIThe feet segmentalmarkers were identified and tracked using the predictedpositions of their centroids based on their velocity at theprevious two frames Cannyrsquos edge operator [28] was usedto obtain the silhouette (Figure 1) and the segmental markerscontours

23 Cycle Segmentation and Gait Parameters DeterminationHeel strike and toe off events were automatically estimatedusing a method originally developed for marker-based sys-tems [29] and adapted to our ML method Per each timeframe the centroids of the pelvis and both feet were deter-mined The gait events (heel strike and toe off) instants weredetermined when the maximum horizontal relative distancebetween the pelvis and the foot centroids is achieved

An expert operator manually identified the same gaitevents using the video footage and the heel and toe 3Dmarker trajectories as reference The following spatial andtemporal parameters were then calculated for both ML andmarker-based data cadence walking speed stride time andstride length The estimated parameters were compared forvalidation purposes

24 Model Calibration A subject-specific multisegmentalmodel of the lower limb was used to track the segments andto compute the relevant joint kinematics The model is madeof four segments (foot tibia femur and pelvis) connected byhinges The position of the following anatomical landmarkswasmanually identified by the operator in the static referenceimage lateral malleolus (LM) lateral femoral epicondyle(LE) and greater trochanter (GT) (Figure 2(a))

The foot model template was defined as the posteriorhalf of the foot segmental marker contour The anatomicalcoordinate system (CSA) of the foot was defined on thefoot model template with the positive 119909-axis coincidentwith the line fitting the lower-posterior contour and orientedtowards the toes The origin was made to coincide with themost posterior point of the foot segmental marker contourA technical coordinate system (CST) was defined with the119909-axis coinciding with the corresponding axis of the CSIand centred in LM (Figure 2(b)) The transformation matrixbetween the CST andCSA of the foot was computed fCSATfCST

The CSA of the tibia was defined with the 119910-axis joiningLM with LE (origin on LM) The tibia model templatewas defined based on ten reference points identified on the


(a) (b) (c)


PL

GT

LE

LMrshk

rtik

(a)

LMCST

CSA

(b)

PLfirstPPfirst

PLlastPPlast

(c)



119896 119896 = 1 10)


119896 119896 = 1 6)




Δ =

1003816100381610038161003816119889119909last minus 119889119909first1003816100381610038161003816

119872 (1)




119889119909119894= 119889119909first + (Δ sdot 119894) (2)



119894)

PL119894= PP119894+ 119889119909119894 (3)





Contourarc

LM

119858k1198490k









0 Due to
















Stride time(s) 001 1 001 1 002 2

Stridelength (m) 003 3 004 3 005 3


3 Results
















RMSD(deg) 119877

2 11986001198601


2 11986001198601


2 11986001198601


2 11986001198601


Slow

Fast

Comfortable

30

20

10

0

0

00

Exte

nsio

nFl

exio

n

10050

0 100500 10050

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

(deg

)

Exte

nsio

nFl

exio

n(d

eg)

Exte

nsio

nFl

exio

n(d

eg)


Gait cycle ()



Slow

Fast

Comfortable

Flex

ion

0 10050

(deg

)

Gait cycle ()

0 10050


Gait cycle ()

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

Flex

ion

(deg

)

Flex

ion

(deg

)

Exte

nsio

n

Exte

nsio

n

Exte

nsio

n





Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5






Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















(a) (b) (c)


PL

GT

LE

LMrshk

rtik

(a)

LMCST

CSA

(b)

PLfirstPPfirst

PLlastPPlast

(c)



119896 119896 = 1 10)


119896 119896 = 1 6)




Δ =

1003816100381610038161003816119889119909last minus 119889119909first1003816100381610038161003816

119872 (1)




119889119909119894= 119889119909first + (Δ sdot 119894) (2)



119894)

PL119894= PP119894+ 119889119909119894 (3)





Contourarc

LM

119858k1198490k









0 Due to
















Stride time(s) 001 1 001 1 002 2

Stridelength (m) 003 3 004 3 005 3


3 Results
















RMSD(deg) 119877

2 11986001198601


2 11986001198601


2 11986001198601


2 11986001198601


Slow

Fast

Comfortable

30

20

10

0

0

00

Exte

nsio

nFl

exio

n

10050

0 100500 10050

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

(deg

)

Exte

nsio

nFl

exio

n(d

eg)

Exte

nsio

nFl

exio

n(d

eg)


Gait cycle ()



Slow

Fast

Comfortable

Flex

ion

0 10050

(deg

)

Gait cycle ()

0 10050


Gait cycle ()

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

Flex

ion

(deg

)

Flex

ion

(deg

)

Exte

nsio

n

Exte

nsio

n

Exte

nsio

n





Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5






Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of


















Δ =

1003816100381610038161003816119889119909last minus 119889119909first1003816100381610038161003816

119872 (1)




119889119909119894= 119889119909first + (Δ sdot 119894) (2)



119894)

PL119894= PP119894+ 119889119909119894 (3)





Contourarc

LM

119858k1198490k









0 Due to
















Stride time(s) 001 1 001 1 002 2

Stridelength (m) 003 3 004 3 005 3


3 Results
















RMSD(deg) 119877

2 11986001198601


2 11986001198601


2 11986001198601


2 11986001198601


Slow

Fast

Comfortable

30

20

10

0

0

00

Exte

nsio

nFl

exio

n

10050

0 100500 10050

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

(deg

)

Exte

nsio

nFl

exio

n(d

eg)

Exte

nsio

nFl

exio

n(d

eg)


Gait cycle ()



Slow

Fast

Comfortable

Flex

ion

0 10050

(deg

)

Gait cycle ()

0 10050


Gait cycle ()

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

Flex

ion

(deg

)

Flex

ion

(deg

)

Exte

nsio

n

Exte

nsio

n

Exte

nsio

n





Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5






Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of





















Stride time(s) 001 1 001 1 002 2

Stridelength (m) 003 3 004 3 005 3


3 Results
















RMSD(deg) 119877

2 11986001198601


2 11986001198601


2 11986001198601


2 11986001198601


Slow

Fast

Comfortable

30

20

10

0

0

00

Exte

nsio

nFl

exio

n

10050

0 100500 10050

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

(deg

)

Exte

nsio

nFl

exio

n(d

eg)

Exte

nsio

nFl

exio

n(d

eg)


Gait cycle ()



Slow

Fast

Comfortable

Flex

ion

0 10050

(deg

)

Gait cycle ()

0 10050


Gait cycle ()

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

Flex

ion

(deg

)

Flex

ion

(deg

)

Exte

nsio

n

Exte

nsio

n

Exte

nsio

n





Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5






Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of



















RMSD(deg) 119877

2 11986001198601


2 11986001198601


2 11986001198601


2 11986001198601


Slow

Fast

Comfortable

30

20

10

0

0

00

Exte

nsio

nFl

exio

n

10050

0 100500 10050

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

30

20

10

minus10

minus20

minus30

minus40

(deg

)

Exte

nsio

nFl

exio

n(d

eg)

Exte

nsio

nFl

exio

n(d

eg)


Gait cycle ()



Slow

Fast

Comfortable

Flex

ion

0 10050

(deg

)

Gait cycle ()

0 10050


Gait cycle ()

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

Flex

ion

(deg

)

Flex

ion

(deg

)

Exte

nsio

n

Exte

nsio

n

Exte

nsio

n





Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5






Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















Slow

Fast

Comfortable

Flex

ion

0 10050

(deg

)

Gait cycle ()

0 10050


Gait cycle ()

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

70

60

40

50

30

20

10

0

minus10

Flex

ion

(deg

)

Flex

ion

(deg

)

Exte

nsio

n

Exte

nsio

n

Exte

nsio

n





Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5






Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















Slow

Fast

Comfortable

0 10050

Plan

tar

Dor

siflex

ion

(deg

)


Gait cycle ()

0 10050

Gait cycle ()

15

10

minus10

minus25

minus20

minus15

0

5

0

5

Plan

tar

Dor

siflex

ion

(deg

)

Plan

tar

Dor

siflex

ion

(deg

)

minus5

15

10

minus10

minus25

minus20

minus15

minus5

0

5

15

10

minus10

minus25

minus20

minus15

minus5






Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















Slow

Fast

Comfortable

0 10050

0 10050

(deg

)

Gait cycle ()

Gait cycle ()

0 10050

Gait cycle ()

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6

Ant

erio

rPo

sterio

r

(deg

)

8

6

4

2

0

minus4

minus6Ant

erio

rPo

sterio

r

minus2 minus2

minus2










References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of






















References










































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of







































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















research article a 2d markerless gait analysis methodology:...

Documents