Gait assessment using the Microsoft Xbox One Kinect: Concurrent validity and inter-day reliability of spatiotemporal and kinematic variables

Journal of Biomechanics 48 (2015) 2166–2170

Contents lists available at ScienceDirect

Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com

Short Communication

Gait assessment using the Microsoft Xbox One Kinect: Concurrent validity and inter-day reliability of spatiotemporal and kinematic variables Benjamin F. Mentiplay a, Luke G. Perraton a, Kelly J. Bower a, Yong-Hao Pua b, Rebekah McGaw a, Sophie Heywood a, Ross A. Clark a,n a b

School of Exercise Science, Australian Catholic University, Melbourne, Australia Department of Physiotherapy, Singapore General Hospital, Singapore

art ic l e i nf o

a b s t r a c t

Article history: Accepted 15 May 2015

The revised Xbox One Kinect, also known as the Microsoft Kinect V2 for Windows, includes enhanced hardware which may improve its utility as a gait assessment tool. This study examined the concurrent validity and inter-day reliability of spatiotemporal and kinematic gait parameters estimated using the Kinect V2 automated body tracking system and a criterion reference three-dimensional motion analysis (3DMA) marker-based camera system. Thirty healthy adults performed two testing sessions consisting of comfortable and fast paced walking trials. Spatiotemporal outcome measures related to gait speed, speed variability, step length, width and time, foot swing velocity and medial–lateral and vertical pelvis displacement were examined. Kinematic outcome measures including ankle flexion, knee flexion and adduction and hip flexion were examined. To assess the agreement between Kinect and 3DMA systems, Bland–Altman plots, relative agreement (Pearson's correlation) and overall agreement (concordance correlation coefficients) were determined. Reliability was assessed using intraclass correlation coefficients, Cronbach's alpha and standard error of measurement. The spatiotemporal measurements had consistently excellent (r Z0.75) concurrent validity, with the exception of modest validity for medial– lateral pelvis sway (r ¼0.45–0.46) and fast paced gait speed variability (r ¼ 0.73). In contrast kinematic validity was consistently poor to modest, with all associations between the systems weak (r o0.50). In those measures with acceptable validity, the inter-day reliability was similar between systems. In conclusion, while the Kinect V2 body tracking may not accurately obtain lower body kinematic data, it shows great potential as a tool for measuring spatiotemporal aspects of gait. & 2015 Elsevier Ltd. All rights reserved.

Keywords: Gaming Low-cost technology Walking Biomechanics Physical function Kinect V2 Kinect for Windows

1. Introduction Gait impairments are prevalent in many clinical populations and the elderly. In these groups, simple and clinically feasible measures such as gait speed are predictive of long term outcomes including health-related quality of life, hospitalization and mortality (Purser et al., 2005; Studenski et al., 2011). Equipment that could allow for spatiotemporal and kinematic gait variables to be assessed routinely in clinical practice may assist in the accurate identification and subsequent treatment of individuals at risk of poor outcomes, such as reduced mobility and falls (Lewek et al., n Correspondence to: School of Exercise Science, Faculty of Health Sciences, Australian Catholic University, Fitzroy, Victoria, 3065 Australia. Tel.: þ 61 431737609; fax: þ61 3 9953 3095. E-mail address: [email protected] (R.A. Clark).

http://dx.doi.org/10.1016/j.jbiomech.2015.05.021 0021-9290/& 2015 Elsevier Ltd. All rights reserved.

2014; Wren et al., 2011). However, it is difficult to acquire these gait variables, such as speed variability, ground contact time and pelvic displacement, as these require instrumentation which is typically not accessible to clinicians. Current devices that are capable of accurately measuring spatiotemporal and kinematic gait variables are typically expensive, time consuming and lack portability. Recent developments in video gaming technology may provide a clinically feasible alternative for gait assessment. The Microsoft Kinect is a low-cost gaming device that has shown promise as a clinical assessment tool (Clark et al., 2015; Galna et al., 2014; Vernon et al., 2015). The original version of the Kinect had the potential to assess spatiotemporal and kinematic gait variables, and given its low cost, availability and markerless tracking, it could provide a clinical alternative to laboratory-based multiple camera systems. This first version of the Kinect is a valid tool for assessing spatiotemporal components of gait (Clark et al.,

B.F. Mentiplay et al. / Journal of Biomechanics 48 (2015) 2166–2170

2167

Table 1 Gait events and variables derived from the anatomical landmark data provided by the Microsoft Kinect skeleton tracking algorithm and the three-dimensional motion analysis (3DMA) system, along with the methods of identification used for each system. Variable

Kinect

3DMA

Left foot: The first sample after vertical ground reaction force drops The foot center local velocity minimum immediately preceding the foot center velocity first exceeding a threshold of þ0.1 m/s in the AP below 10 N during transition to the swing phase. Right foot: The toe marker local velocity minimum immediately preceding the toe marplane ker velocity first exceeding a threshold of þ 0.1 m/s in the AP plane Ground contact The ankle center local velocity minimum immediately after the ankle Left foot: The first sample after vertical ground reaction force exceeds center velocity first dropping below a threshold of þ 0.1 m/s in the AP 10 N during initial ground contact. Right foot: The heel marker local velocity minimum immediately after the heel marker velocity first plane dropping below a threshold of þ 0.1 m/s in the AP plane Gait speed (m/s) Mean velocity of the pelvis center in the AP plane in the range from Mean velocity of the sternum marker in the AP plane in the range 3.5 m to 1.5 m from the Kinect V2 camera from 3.5 m to 1.5 m from the Kinect V2 camera Step length (m) Distance between the left and right limb ground contact positions As per the Kinect Step time (s) Time between left and right limb ground contact As per the Kinect Step width (m) Orthogonal distance to the right ankle joint center at the right limb Orthogonal distance to the right calcaneal marker at the right limb ground contact position from a line running through the left calcaneal ground contact position from a line running through the left ankle marker at the consecutive left limb ground contact positions joint center at the consecutive left limb ground contact positions Pelvis ML range (cm) Total range of motion of the hip center landmark in the ML plane in Total range of motion of the pelvis center in the ML plane in the range the range from 3.5 m to 1.5 m from the Kinect V2 camera from 3.5 m to 1.5 m from the Kinect V2 camera Pelvis AP range (cm) Total range of motion of the hip center landmark in the vertical plane Total range of motion of the pelvis center in the vertical plane in the in the range from 3.5 m to 1.5 m from the Kinect V2 camera range from 3.5 m to 1.5 m from the Kinect V2 camera Foot swing velocity (m/s) Peak velocity of the ankle center in the AP plane in the phase As per the Kinect between intra-limb toe-off and ground contact Peak knee flexion-swing Knee flexion excursion of the left knee during the swing phase of the As per the Kinect (deg) left stride As per the Kinect Peak knee flexion-conKnee flexion excursion of the left knee during the initial contact tact (deg) phase (i.e. absorption) of the left ground contact. Restricted to the first 50% of the ground contact. Peak knee adductionPeak knee adduction of the left knee during the entire ground contact As per the Kinect contact (deg) phase of the left ground contact. Total ankle flexion range Ankle flexion excursion of the left ankle during the entire phase of As per the Kinect (deg) the left stride Hip flexion range (deg) Hip flexion excursion of the left hip during the entire phase of the left As per the Kinect stride Toe-off

Note: AP: anterior–posterior; ML: medial–lateral.

2013; Stone and Skubic, 2011), but is typically poor for assessing lower limb kinematics (Pfister et al., 2014). The Microsoft Xbox One Kinect, also known as Kinect V2 for Windows, was released in 2014, with greatly improved depth and image camera hardware. These changes may improve the automated tracking of anatomical landmarks, potentially enhancing the clinical and research utility of this device for examining gait. However, to our knowledge no prior study has examined the validity and reliability of the Kinect V2 and the official software development kit (SDK) body tracking algorithm for the assessment of gait. Therefore, the aim of this study was to assess the concurrent validity and inter-day reliability of this device when assessing spatiotemporal and kinematic parameters of comfortable and fast paced gait.

2. Methods Thirty young, injury-free individuals (age: 22.87 7 5.08 yrs, height: 172.85 79.11 cm, mass: 68.67 79.15 kg, male: 15) volunteered to participate. Participants attended two testing sessions seven days apart (mean: 7 72 days), with one participant unable to attend the second session. This study was approved by the institution's Human Research Ethics Committee and all participants provided written informed consent. Gait trials were performed along a walkway with an embedded force platform (AMTI OR6) located 2.5 m from the Kinect V2, with the Kinect placed in front of the participant to provide a frontal plane view. The platform was used to identify ground contact and toe-off of the left foot during trials. Participants performed gait trials along this walkway starting approximately 8 m from the Kinect, with the distance adjusted to ensure foot contact in the center of the force plate. This distance allowed for steady state gait to be examined. Participants performed gait trial conditions at two different speeds: comfortable and fast paced. One trial of each pace with a successful ground contact was used for analysis to examine the clinical feasibility of the Kinect. Data from the Kinect V2 were obtained at 30 Hz using the previously described body tracking algorithm included in the Microsoft SDK (Shotton et al., 2011). The anatomical landmarks of the left ankle, left knee, spine

base and spine shoulder were used to represent the left ankle, left knee, pelvis center and manubrium respectively. The data for the 3DMA system were acquired at 100 Hz using a nine camera Vicon system and Nexus software V1.8.5 (Vicon, U. K.). Kinematic data from the Kinect for the left ankle, knee and hip were acquired and converted from quaternion to Euler angles (rotation sequence: XYZ) to allow comparison with the 3DMA system. Data acquired from the 3DMA system were modeled using a custom, cluster-based model similar to Collins et al. (2009), with a minimum of four markers per segment and constrained by inverse kinematics in Visual3D to allow no translations between segments. The joint coordinate systems for both the 3DMA and Kinect were kept consistent with the International Society of Biomechanics recommendations (Wu and Cavanagh, 1995). Spline interpolation was used to resample the Kinect data to 100 Hz prior to analysis. Both data sets were loaded into a custom program, filtered using a Daubechies 4 undecimated wavelet 7.5 Hz low-pass filter, then cross-correlated to temporally align the data traces before analysis. The gait event time points of toeoff and ground contact were used to identify phases of the gait cycle. For both the spatiotemporal and kinematic analysis only the right leg step was assessed. This step commenced with the right foot ground contact and ended with subsequent left foot ground contact on the force platform. Kinematics obtained during the ground contact phase of the movement were assessed during the force platform contact of the left foot, which occurred immediately after the end of the right foot step. The kinematic variables derived for this study were ankle flexion, knee flexion and adduction and hip flexion angles. Kinematic analysis was performed for the full gait cycle, and also independently for the ground contact and swing phases for the knee joint, which is often useful in the gait analysis of clinical populations such as those with knee osteoarthritis (Astephen et al., 2008). The spatiotemporal variables derived consisted of regularly reported measures such as gait speed, step length and time, step width, and foot swing velocity. Additional variables derived related to pelvis movement included: gait speed variability, and medial–lateral and vertical pelvis displacement. These variables were obtained using the supervised automated analysis criteria outlined in Table 1.

2.1. Statistical analysis Pearson's (r) and concordance correlation coefficients (rc) were computed to explore the relative and overall agreement, respectively, between the two methods

2168

B.F. Mentiplay et al. / Journal of Biomechanics 48 (2015) 2166–2170

on day 1. Specifically, the Pearson's correlation assesses association irrespective of magnitude differences whereas the concordance coefficient assesses both association and deviations from the line of identity (y¼ x). Standard or regression-based (when proportional bias was detected) Bland–Altman plots with 95% limits of agreement (Ludbrook, 2010) were calculated for all variables (see Supplementary material). Correlations of the difference between scores and the average scores were examined to detect a proportional bias (r4 0.50), which indicated that a regression-based Bland–Altman plot was to be used. Within-device test–retest reliability was assessed using intraclass correlation coefficients (ICC2,1), Cronbach's alpha and standard error of measurement (expressed as a percentage of the mean). Point estimates of the correlation and ICC values were interpreted as: excellent (0.75–1), modest (0.4–0.74), or poor (0–0.39) (Fleiss, 1986). The relationship between distance from the Kinect and anatomical landmark accuracy error was also determined by examining the linearity of the recorded ground contact position values from both devices.

3. Results Mean (7 SD) values for each variable by device and session, with inter-day reliability and intra-day validity results, are provided in Table 2. Excellent relative agreement (r 40.75) for both conditions was observed for gait speed, ground contact time, vertical pelvis center displacement range, foot swing velocity and all step time, length and width measures. Modest relative agreement was observed for fast paced gait speed variability (r ¼0.73) and medial–lateral pelvis sway (r ¼0.45–0.46). Poor relative agreement was observed for all kinematic variables (r o0.4), with the exception of comfortable pace hip flexion range (r ¼0.49). Overall agreement was typically poor to modest, except excellent results were found (rc40.75) for gait velocity, ground contact time and step length at both gait speeds.

Table 2 Mean ( þ SD) values for each outcome measure using the Microsoft Kinect and three-dimensional motion analysis (3DMA) systems. Results for both conditions (comfortable and fast pace gait) and testing days are provided, along with validity (Pearson's correlation for relative agreement (r) and the concordance correlation coefficient for overall agreement (rc)) and reliability (intraclass correlation coefficient with 95% confidence intervals (ICC2,1) for agreement, standard error of measurement expressed as a percentage of the mean (SEM %) and Cronbach's alpha (a)) results. Day 1 Gait parameters Comfortable pace Spatiotemporal variables Gait velocity (m/s) Gait velocity variability (m/s) Ground contact time (s) Step width (m) Step time (s) Step length (m) Pelvis ML range (cm) Pelvis vertical range (cm) Foot swing velocity (m/ s) Kinematic variables Peak knee flexion-swing (deg) Peak knee flexion-contact (deg) Peak knee adductioncontact (deg) Total ankle flexion range (deg) Hip flexion range (deg) Fast paced Spatiotemporal variables Gait velocity (m/s) Gait velocity variability (m/s) Ground contact time (s) Step width (m) Step time (s) Step length (m) Pelvis ML range (cm) Pelvis vertical range (cm) Foot swing velocity (m/ s) Kinematic variables Peak knee flexion-swing (deg) Peak knee flexion-contact (deg) Peak knee adductioncontact (deg) Total ankle flexion range (deg) Hip flexion range (deg)

Day 2

Validity r

Kinect reliability

Kinect

3DMA

Kinect

3DMA

rc

ICC

1.26 þ 0.12 0.19þ 0.05

1.25 þ0.13 0.15 þ0.04

1.30 þ 0.13 0.19þ 0.04

1.28 þ 0.13 0.15þ 0.03

0.99 0.75

0.90 0.00

0.75 (0.53,0.88) 0.76 (0.55,0.88)

0.44þ 0.06 0.13þ 0.03 0.51 þ0.04 0.66þ 0.07 4.39 þ 1.05 4.53 þ 0.96

0.41þ 0.06 0.19 þ0.03 0.51þ 0.04 0.68þ0.06 3.62 þ1.05 3.86 þ0.96

0.45 þ 0.06 0.13þ 0.02 0.51 þ0.04 0.67þ0.06 4.64þ 1.15 4.67þ0.79

0.41þ 0.05 0.20 þ 0.03 0.52 þ 0.04 0.68þ 0.06 3.58 þ 1.04 3.94 þ 0.81

0.91 0.94 0.92 0.90 0.45 0.87

0.90 0.00 0.75 0.13 0.00 0.00

0.03 0.71 0.70 0.87 0.55 0.76

3.84 þ 0.35 4.33þ 0.40 3.94 þ 0.40 4.40 þ 0.38

0.79

0.11

3DMA reliability SEM % a

ICC

SEM % a

4.8 12.9

0.86 0.76 (0.53,0.88) 0.87 0.80 (0.61,0.91)

13.4 12.4 4.3 3.8 16.8 9.8

0.06 0.83 0.83 0.93 0.72 0.87

0.72 (0.49,0.86)

4.8

0.84

(  0.37,0.42) (0.45,0.86) (0.45, 0.85) (0.75,0.94) (0.23,0.77) (0.55,0.89)

5.1 11.9

0.86 0.89

13.0 7.2 4.2 3.4 15.2 10.0

0.34 0.88 0.83 0.92 0.77 0.91

0.66(0.39,0.83)

5.4

0.80

0.21 0.79 0.71 0.85 0.63 0.84

(  0.21,0.56) (0.58,0.90) (0.47,0.86) (0.71,0.93) (0.34,0.81) (0.68,0.92)

30þ 3

67 þ5

31þ 3

66 þ5

 0.05

 0.02

0.85 (0.71,0.93)

3.9

0.92 0.79 (0.57,0.90)

3.4

0.88

26þ 4

16þ 4

26þ 3

18þ5

 0.01

 0.01

0.72 (0.48,0.86)

8.1

0.84 0.58 (0.24,0.79)

16.2

0.73

4þ 3

3þ5

4þ 3

2þ 4

 0.07

0.00

0.69 (0.42,0.84)

41.8

0.81

0.91 (0.81,0.96)

50.0

0.95

40þ 34

33 þ7

36þ 25

34þ 5

0.11

0.01

0.42 (0.02,0.71)

62.8

0.60 0.75 (0.51,0.88)

10.6

0.86

74 þ23

44þ 6

76þ 23

43þ 5

0.49

0.08

 0.10 (  0.57,0.42)

32.6

 0.23 0.55 (0.16,0.80)

9.1

0.71

1.64þ 0.13 0.27 þ0.05

1.64þ 0.12 0.18 þ0.05

1.62 þ 0.13 1.62 þ 0.13 0.26 þ 0.05 0.18þ 0.04

0.96 0.73

0.83 0.00

0.77 (0.54,0.89) 0.67 (0.38,0.84)

3.8 10.6

0.87 0.79 (0.89,0.90) 0.80 0.74 (0.50,0.87)

3.4 14.2

0.88 0.85

0.34 þ 0.07 0.13þ 0.02 0.47þ 0.04 0.80 þ 0.08 3.57þ 1.02 3.90 þ 1.00

0.29 þ0.06 0.20 þ0.03 0.46 þ0.04 0.80 þ0.07 4.58 þ1.28 4.49þ1.04

0.37þ 0.05 0.13þ 0.03 0.46 þ 0.04 0.78 þ 0.07 3.22 þ 1.07 3.95 þ 0.87

0.32 þ 0.06 0.20 þ 0.03 0.46 þ 0.04 0.78 þ 0.07 4.49þ 1.33 4.45 þ 0.98

0.94 0.92 0.88 0.95 0.46 0.81

0.90 0.00 0.06 0.88 0.00 0.00

0.43 0.74 0.87 0.94 0.68 0.82

(0.04,0.71) (0.49,0.87) (0.74,0.94) (0.87,0.97) (0.41,0.84) (0.64,0.91)

15.5 7.8 3.1 2.4 11.3 10.6

0.60 0.85 0.93 0.97 0.81 0.90

0.47 0.78 0.90 0.85 0.64 0.94

(0.07,0.74) (0.58,0.90) (0.80,0.96) (0.69,0.93) (0.35,0.82) (0.88,0.97)

15.1 7.0 2.7 3.4 15.0 6.1

0.64 0.88 0.95 0.92 0.78 0.97

4.97 þ0.43

5.46 þ0.40 4.93 þ 0.52 5.40 þ 0.47

0.88

0.18

0.58 (0.26,0.79)

5.6

0.74

0.68 (0.40,0.84)

4.1

0.81

33þ 4

69þ5

33þ 3

67þ 5

 0.30

 0.09

0.75 (0.52,0.88)

6.1

0.86 0.63 (0.31,0.82)

4.4

0.77

28þ 4

20þ4

29þ 6

18þ5

 0.17

 0.19

0.56 (0.22,0.77)

9.5

0.72

0.57 (0.22,0.79)

13.1

0.73

5þ 4

3þ5

5þ 3

2þ 4

0.23

0.01

0.38 (0.00,0.67)

63.0

0.55 0.88 (0.74,0.95)

57.7

0.94

41þ 28

32þ5

44 þ33

32 þ6

 0.26

 0.02

0.44 (  0.04,0.75)

51.1

0.61 0.68 (0.38,0.85)

8.8

0.81

75þ 35

53þ7

66 þ 15

54þ 8

0.17

0.01

10.7

0.50

Note: see Supplementary material for description of outliers removed during analyses.

0.23 (  0.40,0.71) 40.9

0.38 0.34 (  0.05,0.64)

B.F. Mentiplay et al. / Journal of Biomechanics 48 (2015) 2166–2170

The reliability of all spatiotemporal results was similar between devices. Reliability of most kinematic variables was modest to poor for the Kinect V2 with the exception of peak knee flexion angle during the swing phase. The reliability was mostly excellent to modest for the 3DMA system, with the worst results for hip flexion range of motion and knee flexion during the ground contact phase. Results were typically similar between the two gait speed conditions. The linearity assessment for the accuracy of landmark detection with respect to distance from the Kinect V2 was excellent, with no noticeable effect of distance.

4. Discussion The present study found that the Kinect V2 used with the automated body tracking algorithm is a promising tool for assessing spatiotemporal variables but not lower body kinematics during comfortable and fast paced gait in young healthy people. There was typically an excellent relative association between spatiotemporal values collected using both devices. For the outcome measures with excellent concurrent validity, reliability results were consistent between devices. However, caution must be taken when interpreting results derived from the Kinect V2 in the context of established normative data. The concordance correlation outcomes revealed that, in many cases, the results cannot be considered interchangeable with those obtained using a 3DMA system. This is not surprising given that the anatomical landmark identification differs between protocols. While this finding is important, it does not discount the potential use of this system for clinical and field-based assessment of gait. An important finding of this study was the poor kinematic validity results. The position chosen for the Kinect V2 camera in this study may have had a negative effect on the kinematic results obtained. From our piloting experience, positioning the Kinect directly in front of the person walking towards it provides the optimal spatiotemporal data, with the strong validity results obtained in this study showing the capabilities of the Kinect when used in this position. However, this position may not provide the best placement for assessing joint angles of the lower body. Prior research using the first iteration of the Kinect showed that lower limb joint kinematics could be obtained with modest accuracy if the Kinect was positioned at a 45° angle to the longitudinal walking plane (Pfister et al., 2014). This was not done for two primary reasons. Firstly, we chose a frontal plane viewpoint to maximize the accuracy of the spatiotemporal results whilst still being able to obtain kinematic data from both lower limbs and the pelvis in a clinically feasible single trial. This may also be a reason for many of our lower than expected reliability values for the 3DMA system, as prior research typically averages across multiple trials (McGinley et al., 2009). Secondly, pilot testing of the kinematic data obtained from different angles did not reveal any positions which resulted in particularly accurate and sensitive data. This is especially important in the context of the inter-individual range of motion differences observed during gait, with small variability in healthy populations for joints such as the knee (Brinkmann and Perry, 1985). From our experience with the Kinect V2, and the results of this and prior gait validation studies using the first iteration Kinect (Pfister et al., 2014), the variability between healthy participants in kinematic data is too small to be detectable given the joint angle error obtained. However, future research may consider examining different Kinect V2 placement positions to optimize the accuracy of kinematic data, or implement marker-based tracking to improve the accuracy of landmark identification. The findings from this study may have differed if the assessment was undertaken in a clinical population. For example, gait

2169

speed variability and step width would likely be more pronounced and have greater inter-individual variability in people with stroke or balance deficits (Chen et al., 2005; Patterson et al., 2008). This increased heterogeneity would potentially lead to higher reliability estimates and stronger correlations between the systems. The spatiotemporal gait variables derived from the Kinect may assist clinicians to detect specific gait deficits which may be amenable via targeted treatment approaches. Furthermore, these variables may be more sensitive to subtle changes in gait patterns than the traditional gross measure of gait speed, and could be useful for predicting functional outcomes. In conclusion, despite its inability to accurately assess lower body kinematics during gait, the Kinect V2 has great potential as a low-cost, easily implemented device for assessing spatiotemporal components of gait. Future research should examine its utility in a clinical setting in patient populations who are likely to have gait impairments.

Conflict of interest statement The authors report no conflict of interest.

Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jbiomech.2015.05. 021.

References Astephen, J.L., Deluzio, K.J., Caldwell, G.E., Dunbar, M.J., 2008. Biomechanical changes at the hip, knee, and ankle joints during gait are associated with knee osteoarthritis severity. J. Orthop. Res. 26, 332–341. Brinkmann, J.R., Perry, J., 1985. Rate and range of knee motion during ambulation in healthy and arthritic subjects. Phys. Ther. 65, 1055–1060. Chen, G., Patten, C., Kothari, D.H., Zajac, F.E., 2005. Gait differences between individuals with post-stroke hemiparesis and non-disabled controls at matched speeds. Gait Posture 22, 51–56. Clark, R.A., Bower, K.J., Mentiplay, B.F., Paterson, K., Pua, Y.H., 2013. Concurrent validity of the Microsoft Kinect for assessment of spatiotemporal gait variables. J. Biomech. 46, 2722–2725. Clark, R.A., Vernon, S., Mentiplay, B.F., Miller, K.J., McGinely, J.L., Pua, Y.H., Paterson, K., Bower, K.J., 2015. Instrumenting gait assessment using the Kinect in people living with stroke: Reliability and association with balance tests. J. Neuroeng. Rehabil. 12, 1–9. Collins, T.D., Ghoussayni, S.N., Ewins, D.J., Kent, J.A., 2009. A six degrees-of-freedom marker set for gait analysis: repeatability and comparison with a modified Helen Hayes set. Gait Posture 30, 173–180. Fleiss, J.L., 1986. The Design and Analysis of Clincal Experiments. John Wiley & Sons, New York, USA. Galna, B., Barry, G., Jackson, D., Mhiripiri, D., Olivier, P., Rochester, L., 2014. Accuracy of the Microsoft Kinect sensor for measuring movement in people with Parkinson's disease. Gait Posture 39, 1062–1068. Lewek, M.D., Bradley, C.E., Wutzke, C.J., Zinder, S.M., 2014. The relationship between spatiotemporal gait asymmetry and balance in individuals with chronic stroke. J. Appl. Biomech. 30, 31–36. Ludbrook, J., 2010. Confidence in Altman–Bland plots: a critical review of the method of differences. Clin. Exp. Pharmacol. Physiol. 37, 143–149. McGinley, J.L., Baker, R., Wolfe, R., Morris, M.E., 2009. The reliability of threedimensional kinematic gait measurements: A systematic review. Gait Posture 29, 360–369. Patterson, K.K., Parafianowicz, I., Danells, C.J., Closson, V., Verrier, M.C., Staines, W. R., Black, S.E., McIlroy, W.E., 2008. Gait asymmetry in community-ambulating stroke survivors. Arch. Phys. Med. Rehabil. 89, 304–310. Pfister, A., West, A.M., Bronner, S., Noah, J.A., 2014. Comparative abilities of Microsoft Kinect and Vicon 3D motion capture for gait analysis. J. Med. Eng. Technol. 38, 274–280. Purser, J.L., Weinberger, M., Cohen, H.J., Pieper, C.F., Morey, M.C., Li, T., Williams, G. R., Lapuerta, P., 2005. Walking speed predicts health status and hospital costs for frail elderly male veterans. J. Rehabil. Res. Dev. 42, 535–545. Shotton, J., Fitzgibbon, A., Cook, M., Sharp, T., Finocchio, M., Moore, R., Kipman, A., Blake, A., 2011. Real-time human pose recognition in parts from single depth

2170

B.F. Mentiplay et al. / Journal of Biomechanics 48 (2015) 2166–2170

images. In: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Stone, E., Skubic, M., 2011. Evaluation of an inexpensive depth camera for in-home gait assessment. J. Ambient Intell. Smart Environ. 3, 349–361. Studenski, S., Perera, S., Patel, K., Rosano, C., Faulkner, K., Inzitari, M., Brach, J., Chandler, J., Cawthon, P., Connor, E.B., Nevitt, M., Visser, M., Kritchevsky, S., Badinelli, S., Harris, T., Newman, A.B., Cauley, J., Ferrucci, L., Guralnik, J., 2011. Gait speed and survival in older adults. J. Am. Med. Assoc. 305, 50–58.

Vernon, S., Paterson, K., Bower, K., McGinley, J., Miller, K., Pua, Y.H., Clark, R.A., 2015. Quantifying individual components of the timed up and go using the kinect in people living with stroke. Neurorehabilit. Neural Repair 29, 48–53. Wren, T.A.L., Gorton, G.E., Õunpuu, S., Tucker, C.A., 2011. Efficacy of clinical gait analysis: a systematic review. Gait Posture 34, 149–153. Wu, G., Cavanagh, P.R., 1995. ISB recommendations for standardization in the reporting of kinematic data. J. Biomech. 28, 1257–1261.