Title: Evaluation of a portable markerless finger position capture device: accuracy of the Leap 1

Motion controller in healthy adults 2

Authors: James Y Tung1, Tea Lulic2, Dave A Gonzalez2, Johnathan Tran2, Clark R Dickerson2, 3

Eric A Roy2 4

5

Affiliations: 1Department of Mechanical and Mechatronics Engineering; 2Department of 6

Kinesiology, University of Waterloo, Waterloo, ON, N2L 3G1 7

8

Corresponding author: James Y Tung (PhD, PEng), Department of Mechanical and Mechatronics 9

Engineering, 200 University Ave. W, Waterloo, ON, N2L 3G1 Canada, Tel. 519-888-4567 10

x33445, Email. james.tung@uwaterloo.ca 11

12

Short title: Leap Motion controller finger position accuracy in healthy adults 13

14

Keywords: motor control, upper limb, kinematics, motion capture 15

16

Word count (not incl figures, tables, references, and abstract): 3,257 17

18

Abstract 19

Although motion analysis is frequently employed in upper limb motor assessment (e.g., 20

visually-guided reaching), they are resource-intensive and limited to laboratory settings. This study 21

evaluated the reliability and accuracy of a new markerless motion capture device, the Leap Motion 22

controller, to measure finger position. Testing conditions that influence reliability and agreement 23

between the Leap and a research-grade motion capture system were examined. Nine healthy young 24

adults pointed to 1 of 15 targets on a computer screen under 2 conditions: 1) touching the target 25

(Touch) and 2) 4 cm away from the target (No-Touch). Leap data was compared to an Optotrak 26

marker attached to the index finger. Across all trials, root mean square (RMS) error of the Leap 27

system was 17.30±9.56 mm (mean±SD), sampled at 65.47±21.53 Hz. The % viable trials and mean 28

sampling rate were significantly lower in the Touch condition (44% versus 64%, p < 0.001; 29

52.02±2.93 versus 73.98±4.48 Hz, p=0.003). While linear correlations were high (horizontal: r2 = 30

0.995, vertical r2 = 0.945), the limits of agreement were large (horizontal: -22.02 to +26.80 mm, 31

vertical: -29.41 to +30.14 mm). While not as precise as more sophisticated optical motion capture 32

systems, the Leap Motion controller is sufficiently reliable for measuring motor performance in 33

pointing tasks that do not require high positional accuracy (e.g., reaction time, Fitt’s, trails, 34

bimanual coordination). 35

36

37

1.0 Introduction 38

Assessment of upper limb impairment and function is critical for diagnosis, treatment, and 39

rehabilitation of musculoskeletal and neural disorders affecting the hand and digits. Hand and digit 40

function are commonly impaired following stroke and are among the least likely to fully recover 41

with conventional rehabilitation (Schieber et al. 2009). Assessment of upper limb function using 42

clinical instruments, such as the Nine Hole Peg Test (Oxford Grice et al. 2003) and Chedoke-43

McMaster Stroke Assessment(CMSA) scale (Gowland et al. 1993) is fundamental to developing 44

rehabilitation goals, monitoring recovery, and adapting treatment based on progress. While such 45

assessments are widely used, they may not sufficiently measure small, but clinically significant 46

changes which may better infer the state of recovery (Gladstone et al. 2002). Further, these 47

assessments require valuable time by trained therapists and are typically limited to episodic 48

intervals. As a result, there may be suboptimal feedback to clinicians and patients in the 49

rehabilitation process, leading to hindered or limited recovery. 50

In contrast, commercial motion tracking systems (e.g., Vicon, Optotrak) are capable of 51

collecting quantitative three-dimensional kinematics across a wide range of tasks (e.g., pointing, 52

reaching, grasping). These systems allow generation of in-depth kinematic measures, such as limb 53

positions and velocities accurate to sub-millimeter level (Windolf et al. 2008), to assess functional 54

capabilities. For example, kinematics of reaching and pointing tasks have been used to assess upper 55

limb impairment and recovery following stroke (Levin 1996; Coderre et al. 2010; Kitago et al. 56

2013; Kamper et al. 2002). Although optical motion analysis systems (e.g., Optitrak, Optotrak, 57

Vicon) are frequently employed in research, they are expensive (ranging from $7,500 to $415,000 58

USD in recent comparisons (Thewlis et al. 2013; Carse et al. 2013)), require training, setup time 59

for affixing markers and calibration, are sensitive to lighting conditions, need a line of sight, and 60

are typically only used in dedicated laboratory settings. These limitations present major barriers to 61

the wide clinical use of data from these motion capture systems (Rosenhahn et al. 2008) and to the 62

emerging development of at-home rehabilitation technologies (Borghese et al. 2013; Slijper et al. 63

2014; Pastor et al. 2012). 64

Advances in sensor technologies have increased accessibility and portability of upper limb 65

kinematics assessments. Inertial measurement units (IMUs), incorporating combinations of 66

accelerometer, gyroscope, and magnetometer sensors, have gained increasing use to measure 67

physical activity, gait and posture, and gross limb movements (Gebruers et al. 2010). While the 68

use of IMUs are validated for measuring the duration and intensity of upper limb movements, their 69

use in capturing the quality of upper limb movements, such as kinematics, smoothness, and 70

movement type (e.g., voluntary vs. unintentional), remains limited (Lemmens et al. 2014). 71

Furthermore, IMU techniques are susceptible to artifacts due to nearby metal objects (Brodie et al. 72

2008). While instrumented gloves (Kortier et al. 2012; Simone & Kamper 2005) for the capture of 73

hand and digit kinematics exist, their use is limited by difficulties in donning and removing the 74

gloves, especially in patients with functional impairments. A number of robotic systems, such as 75

the KinARM (Scott 1999), HapticMaster (Lemmens et al. 2014), and BONES (Milot et al. 2014) 76

systems, have also been developed to facilitate delivery of rehabilitation training. While these 77

systems are capable of high-accuracy position measurements, their utility is limited by relatively 78

high costs and long setup time. 79

Low-cost markerless optical systems have been examined as potential upper limb 80

assessment tools. The Kinect sensor (Microsoft Inc., Redmond, WA), which projects and detects 81

changes in structured infrared light patterns (i.e., dot pattern) to track the movement of body 82

segments in 3D, represents a potential alternative to measure gross body movement in a clinical 83

setting (Kurillo et al. 2013; Mobini et al. 2014). However, the spatial resolution of the Kinect limits 84

the accuracy of measuring movements of small segments (Khoshelham & Elberink 85

2012)(Khoshelham & Elberink 2012), such as fingers. To improve accuracy of estimating pointing 86

position, new techniques using proximal joint (i.e., wrist, elbow) angles to calculate pointing 87

vectors are under development (Fürntratt & Neuschmied 2014)(Fürntratt & Neuschmied 2014). 88

According to company reports, the Leap Motion controller is comprised of two 89

monochromatic stereo cameras and 3 infrared (wavelength = 850 nm) LEDs to reconstruct the 90

scene in 3D and track hand and finger position (Colgan 2014). The system’s low cost (retail 91

$80USD) and minimal set-up time, coupled with initial reports of high accuracy, indicates its 92

potential for assessment of finger movement in three-dimensional space. Independent evaluations 93

using an industrial robot with a pen-like tip reported mean squared error between the tip position 94

and position measured by the Leap system was <0.2 mm in static trials, and <2.5 mm in dynamic 95

trials (Weichert et al. 2013). In an evaluation using a plastic arm model, Guna et al. found the Leap 96

system accurate to <0.5mm in static conditions (Guna et al. 2014), further indicating initial promise 97

for clinical assessment. Furthermore, a recent comparison reported a considerably lower end-to-98

end latency of the Leap (85 ms) compared to the Kinect (170 ms) (Brown et al. 2014)(Brown et 99

al. 2014). Considering the variability in human performance compared to robotic arms is well-100

documented (e.g., van der Smagt et al. 2009), an important next step prior to clinical use is to 101

evaluate the system’s validity and reliability with human participants. 102

The purpose of this study was to evaluate the Leap system for measuring finger position 103

for clinically relevant neuromotor assessments, such as reaching and pointing, in human 104

participants. Specifically, the objectives of this study were to: 1) examine testing conditions that 105

influence the system’s reliability to measure finger position, particularly the effect of touching 106

external objects (e.g., targets on computer screen), and finger location relative to the sensor’s field 107

of view (FOV) and 2) evaluate the validity and limits of agreement between the Leap and a 108

validated research-grade optical motion capture systems. It was hypothesized that reliability of the 109

Leap system to measure finger position, indicated by % of viable trials and mean sampling rate, 110

would be compromised when touching visual stimuli presented on a computer screen compared to 111

a condition where the finger was not touching the screen. We also hypothesized that reliability will 112

be high for targets located in the reported 150 degree field of view and reduced at the margins. 113

With regards to validity and limits of agreement, it was hypothesized that correlation between the 114

Leap and Optotrak systems, measured by r2, and accuracy, measured by root mean square (RMS) 115

error, would be sufficient for reach and point tasks typically used in neuromotor assessments based 116

on previously reported upper limb assessments following stroke (e.g., Coderre et al., 2010; Kamper 117

et al., 2002; Levin, 1996). 118

119

2.0 Methods 120

2.1 Participants 121

Nine healthy right hand dominant younger adults between 18 and 30 years of age (6 122

females; 3 males) participated in this study. Participants did not report any musculoskeletal or 123

neurological disorders, and provided informed consent. This study was approved by the Office of 124

Research Ethics at the University of Waterloo. 125

2.2 Apparatus 126

A desktop computer was used to display a visual stimulus chosen randomly from 15 square 127

targets via E-prime (v 1.2, Psychology Software Tools, Inc., Sharpsberg, PA, USA). Figure 1 128

depicts the target locations and dimensions of the experimental setup. An Infrared Emitting Diode 129

(Ired) was placed on the distal phalange of the index finger and an Optotrak 3020 (Northern Digital 130

Inc., Waterloo, ON, Canada) motion capture system was used to record the location of the finger. 131

A Leap Motion controller system (version 1.2.0, Leap Motion Inc., San Francisco, CA, US), 132

connected to a laptop computer (Lenovo G550, Lenovo Inc., Beijing, China) was used to measure 133

the index finger location using a custom-built C# application to collect the Leap data stream. 134

Briefly, three infrared LEDs on the controller emit a pattern of dots which are measured using two 135

IR cameras and synthesized into 3-D positional data using a proprietary algorithm (Leap Motion 136

Inc., San Francisco, CA, US). Based on recommended placement in the user guide, the Leap sensor 137

was situated 7.5 cm in front of the display monitor and 26 cm from edge of the table (see Figure 138

1), with axes aligned with the Optotrak coordinate frame. 139

2.3 Procedure 140

From an initial starting position, participants were instructed to move their finger to 1 of 141

15 targets presented on the computer monitor (Figure 1) and maintain this position. Participants 142

were asked to perform two variations of this task: 1) touching the monitor (Touch condition) and 143

2) holding their finger 4 cm away from the screen perpendicular to the monitor at the target location 144

(No-Touch condition). In the No-Touch condition, a 4 cm diameter ball was used to control the 145

distance between the finger and monitor and was removed prior to each trial. Once the participants’ 146

fingers were in the final position, three seconds of data were collected using the Optotrak (sampling 147

rate: 120 Hz) and the Leap (sampling rate range: 50 – 120 Hz) systems. Data acquisition was 148

synchronized by simultaneous pressing of mouse buttons by the researchers. While this procedure 149

is not suitable for frame-by-frame comparisons, the reported static positions averaged over a 3 150

second trial duration is largely independent of the small variations in synchronization. 151

Participants performed 3 trials at each of 15 targets and 2 conditions for a total of 90 trials. 152

Target location was randomized for each trial. The Touch/No-Touch conditions were tested in 153

blocks, with order of presentation randomized across participants. After 5 participants were tested 154

for both Touch/No-Touch conditions, it was apparent that the Touch condition produced unreliable 155

results (56% of trials with no data, see Table 1). The remaining 4 participants were tested in the 156

No- Touch condition only. 157

2.4 Data Processing 158

All data was processed using Matlab R13 (Mathworks Inc., Natick, MA, USA). Trials in 159

which the Leap system collected 80 points or less were labeled as ‘No data’. For trials with >80 160

points, mean sampling rate was calculated as the number of points collected over the 3 s trial 161

duration. For anti-aliasing purposes, Leap and Optotrak data were digitally filtered using a dual 162

low-pass Butterworth filter (3rd order), with a cut off of 10Hz. To permit digital filtering, position 163

data collected by the Leap system was resampled at a consistent 60 Hz using a piecewise cubic 164

hermite interpolating polynomial (pchip Matlab function). To compare the Leap and Optotrak 165

systems, the center target (Target 8) was chosen as the reference origin due to the target’s high 166

reliability and consistent sampling rate (see Table 1). The reference origin was calculated as the 167

mean finger position for Target 8 trials for each participant. All position data was referenced to 168

the corresponding participants’ calculated origin. 169

2.5 Analyses 170

To evaluate the influence of the Touch condition on measurement reliability, the proportion 171

of viable trials and mean sampling rates were compared to the No-Touch condition. Reliability 172

was measured by comparing the proportion (%) of viable trials (i.e., 80 or more data points) and 173

mean sampling rate. Accuracy was measured by calculating the mean RMS error of the Leap 174

position compared to the measured Optotrak position. Fisher’s exact test was used to test the null 175

hypothesis that Touch and No-Touch conditions produced the same proportion of viable trials. 176

Similarly, task-dependent difference in mean sampling rate was compared using a paired t-test. To 177

examine the influence of field of view (FOV), the proportion of viable trials from targets at the 178

Outside edges of the grid (Targets 1, 5, 6, 10, 11, and 15) were compared to targets in the Center 179

(Targets 3, 8, and 13). Fisher’s exact test was used to test the null hypothesis that the Outside and 180

Center targets produced the same proportion of viable trials. 181

Validity and limits of agreement between the Leap and Optotrak systems were examined 182

using correlation and Bland-Altman analyses. Separate analyses were conducted for the horizontal 183

and vertical axes. In the correlation analysis, a least-squares regression line was fitted (regress 184

Matlab function). The coefficient of determination (r2) was used as a measure of linear fit. Used 185

extensively to evaluate the limits of agreement between two techniques, Bland-Altman plots were 186

constructed to examine systematic bias and estimate confidence intervals of measurement error. 187

Plots were generated using the mean of the two methods as the abscissa (x-axis) and difference 188

between Leap and Optotrack positions as the ordinate (y-axis) values. The mean errors and 95% 189

confidence intervals for each plane were reported. Considering the negative impact of the Touch 190

condition on reliability, only the No Touch data were examined in the correlation and Bland-191

Altman analyses. 192

193

3.0 Results 194

Across all conditions and targets, the mean (±SD) RMS error of the Leap system was 17.30 195

(±9.56) mm, sampled at 65.47 (±21.53) Hz, with 244/450 (54.2%) viable trials. Table 1 reports 196

descriptive statistics of means and standard deviations of RMS error, sampling rate, and % viable 197

trials by condition and target. 198

3.1 Touching vs No-Touching 199

Touching the computer screen negatively impacted the proportion of trials with viable data 200

(i.e., 80 or more data points) and mean sampling rate. As hypothesized, touching the computer 201

monitor (Touch) reduced the proportion of viable trials compared to the not touching (No Touch) 202

condition (44% versus 64%, p < 0.001). Table 2 reports contingency tables of viable trials by 203

condition. Mean (±SE) sampling rate was also lower in the Touch condition compared to the No 204

Touch condition (52.02±2.93 versus 73.98±4.48 Hz, p=0.003). 205

3.2 Outside vs Center 206

Target location affected the ability to collect finger position data. The proportion of viable 207

trials was greater in the Centre locations compared to the Outside locations (78% versus 60%, 208

p=0.009). However, in comparing the Centre and Outside targets, there was no difference in mean 209

(±SE) sampling rates (65.06±1.94 versus 68.21±1.99 Hz, p = 0.263). 210

3.3 Correlation 211

Considering the influence of the Touch condition on the reliability of the Leap system, only 212

the No Touch condition results are reported in the concurrent validity (i.e., correlation) and limits 213

of agreement (i.e., Bland-Altman) analyses. To evaluate the concurrent validity of the Leap and 214

Optotrak systems, the linear correlation between measured finger positions was examined. Overall, 215

our results show high correlation and coefficient of determination in both the horizontal (r = 0.953, 216

r2 = 0.995, Fig. 2 left) and vertical (r=0.991, r2 = 0.945, Fig. 2 right) directions. 217

3.3 Bland-Altman 218

To examine the limits of agreement between the Leap and Optotrak systems, Bland-Altman 219

plots were generated for the horizontal (Fig. 3, left) and vertical (Fig. 3, right) directions. Similar 220

to the correlation analysis, only the No Touch condition data were used due to poorer reliability in 221

the Touch condition. For the horizontal axis, the mean difference between the two systems was 222

+2.39 mm (Fig. 3, left, solid line), with lower and higher 95% limits of agreement between -22.02 223

and +26.80 mm (Fig. 3, left, dashed lines). In the vertical axis, the mean difference was +0.37 mm 224

(Fig. 3, right, solid line), with limits of agreement between -29.41 and +30.14 mm (Fig. 3, right, 225

dashed lines). 226

227

4.0 Discussion 228

To improve care and facilitate recovery of the hand and its digits, such as paresis of the 229

hand and digits following stroke, there is a need to develop new portable tools for rapid clinic- and 230

home-based assessment. Towards this goal, the purpose of this study was to evaluate the utility 231

of a new markerless kinematics capture device, the Leap Motion Controller, to measure static 232

finger position in a group of healthy younger adults. Comparison of the finger positions measured 233

by the Leap system and a research-grade optoelectronic motion capture system demonstrate that 234

the Leap system: 1) is less reliable when touching visual stimuli presented on a monitor, 2) is 235

inconsistent near the ranges of the device’s field of view (FOV), 3) provides valid measures of 236

finger position indicated by high coefficients of determination (r2), and 4) has a relatively high 237

mean RMS error and limits of agreement. While not as precise as more sophisticated systems, the 238

overall reliability and accuracy of Leap system is sufficient for pointing tasks that do not require 239

high positional accuracy measures over a limited functional space and when the finger does not 240

contact an object, such as a computer screen. However, the RMS error and limits of agreement 241

suggest that the Leap system is not precise enough to measure impaired path deviations in motor-242

impaired populations. 243

As hypothesized, the reliability of the Leap system was significantly reduced when 244

touching visual stimuli and approaching the periphery of the range of the field of view, indicated 245

by a lowered proportion of viable trials and mean sampling rate. In the current study, touching 246

targets resulted in a 20% reduction in viable trials and 21% decrease in mean sampling rate 247

compared to the No Touch condition. This may be attributable to merging of the finger and screen 248

by the Leap system, making differentiation of the two impossible. This limitation restricts the 249

ability to employ assessment tasks that require tactile interaction with buttons (e.g., reaction time), 250

screens (e.g., Trail Making Test), or objects (e.g., blocks, tools). In the evaluation of reliability, 251

outer targets demonstrated 18% less viable trials compared with targets along the center line, with 252

no difference in sampling rate. This reduction may be attributed to occlusions or merging 253

associated with the position and orientation of the hands and fingers relative to the sensor. At wider 254

angles, the pointing finger was more likely to be obstructed from view by individual hand and 255

finger orientations and anatomical dimensions. 256

The Leap system demonstrated a high degree of correlation and thus concurrent validity 257

with a validated Optotrak system (r2 = 0.995 and 0.945 in horizontal and vertical axes, 258

respectively). However, the mean RMS error (17.30 mm) in the current study is considerably 259

higher compared to previous evaluations of the Leap accuracy. Weichert et al.’s study (2013) 260

reported position error was less than 0.17 mm, and Guna et al. (2014) found a maximum standard 261

deviation of 0.49 mm across a similar space. The large differences in error are likely attributable 262

to variability in human data in the current study compared to robotic actuators (van der Smagt et 263

al. 2009). While the previous evaluations document the baseline capabilities of the technology 264

under ideal conditions, the current study better reflects conditions for clinical use. 265

The limits of agreement and mean RMS error are relatively large compared to clinically-266

relevant measures of path deviation and active range of motion in visually-guided reaching tasks. 267

In stroke studies, Levin reported a mean difference of 12.5 mm in end-point accuracy comparing 268

impaired (i.e., affected) and unimpaired arms in a rapid pointing task (Levin 1996). Schaefer et 269

al. found less than 15 mm differences in constant error between healthy controls and stroke 270

survivors (Schaefer et al. 2007; Schaefer et al. 2009). Considering the limits of agreement and 271

RMS error found in the current study are greater than these reported effect sizes, we do not 272

recommend using the Leap system to measure positional accuracy in rapid pointing tasks. 273

Despite its shortcomings in positional accuracy and precision, the Leap may be suitable for 274

a subset of motor assessment tasks, such as Trail-Making or bimanual coordination, with 275

appropriate design modifications. Adapting tasks to use non-contact targets or cues, such as 276

auditory or virtual feedback, are appropriate modifications to address contacting object limitations 277

of the Leap system. An adapted Trail-Making task (Reitan 1958), for example, could 278

simultaneously display finger position and targets on a screen to assess motor planning 279

capabilities. Auditory cues, such as a metronome, may be used to cue bimanual motor coordination 280

tasks such as circle-drawing (Bangert et al. 2010). Considering the poorer accuracy of the Leap at 281

the limits of the evaluation space, we recommend designing tasks in the center of the functional 282

volume. In particular, we recommend maintaining a height of approximately 20 cm above the 283

sensor in light of the poor reliability results at lower heights. With regards to sensitivity in the 284

depth dimension, the current study is limited to examining Leap sensitivity across varying 285

positions in the vertical and horizontal planes. Sensitivity to sensor placement in the depth 286

dimension remains to be examined. 287

Given the Leap Motion costs significantly less relative to laboratory motion tracking 288

devices, the findings in this study are promising despite the reduced accuracy. As a portable tool, 289

the sensor can be used to collect data in a patient’s home, integrated as a feedback system, and as 290

a clinical rehabilitation tool. The present findings were executed on the first commercially-291

available version of the Leap Motion sensor (version 1.2.0). Since the company is working towards 292

improving the tracking capabilities of the device, the results presented in this study may differ 293

from the findings in future studies, which use later, more refined device versions. Furthermore, 294

research is needed to examine dynamic response characteristics to evaluate the Leap’s ability to 295

measure additional kinematic variables (i.e., velocities and acceleration). 296

297

Conclusion 298

The LEAP sensor successfully measured relative 2-D coordinates of the index finger with 299

the highest correlations achieved in the left-right and up-down axes in the No Touch condition, 300

using the Optotrak system as a gold standard. Overall, the Leap system’s accuracy for finger 301

location was 17.3±9.56SD mm compared to criterion standard optical motion capture system 302

(Optotrak). However, some caution is warranted considering the relatively large limits of 303

agreement (horizontal: -22.0 to 26.8 mm, vertical: -29.4 to 30.1 mm) and limitations with 304

contacting objects. Hence, Leap Motion should provide a portable 3-D motion capture system for 305

performing basic clinical assessments in tasks such as trail-making, Fitts, or simple/choice reaction 306

time. One of the major advantages of Leap sensor is its ability to generate continuous data as the 307

hands are moved in space. Thus, it is capable of collecting a greater quantity and quality of 308

information relative to the conventional physiotherapy tools. 309

310

Acknowledgements 311

We acknowledge the assistance of Leah van Draanen with collection of data for this study. The 312

authors declare no conflicts of interest in association with this study. 313

314

315

References 316

Bangert, A.S. et al., 2010. Bimanual coordination and aging: neurobehavioral implications. 317 Neuropsychologia, 48(4), pp.1165–70. 318

Borghese, N.A. et al., 2013. Computational Intelligence and Game Design for Effective At-319 Home Stroke Rehabilitation. Games for health journal, 2(2), pp.81–88. 320

Brodie, M.A., Walmsley, A. & Page, W., 2008. Dynamic accuracy of inertial measurement units 321

during simple pendulum motion. Computer methods in biomechanics and biomedical 322 engineering, 11(3), pp.235–42. 323

Brown, M.A., Stuerzlinger, W. & Filho, E.J.M., 2014. The performance of un-instrumented in-324 air pointing. In Proceedings of the 2014 Graphics Interface Conference. Canadian 325

Information Processing Society, pp. 59–66. 326

Carse, B. et al., 2013. Affordable clinical gait analysis: an assessment of the marker tracking 327 accuracy of a new low-cost optical 3D motion analysis system. Physiotherapy, 99(4), 328 pp.347–51. 329

Coderre, A.M. et al., 2010. Assessment of upper-limb sensorimotor function of subacute stroke 330 patients using visually guided reaching. Neurorehabilitation and neural repair, 24(6), 331 pp.528–41. 332

Colgan, A., 2014. How does the Leap Motion Controller work? blog.leapmotion.com, 333 p.http://blog.leapmotion.com/hardware–to–software–ho. Available at: 334

http://blog.leapmotion.com/hardware-to-software-how-does-the-leap-motion-controller-335 work/. 336

Fürntratt, H. & Neuschmied, H., 2014. Evaluating Pointing Accuracy on Kinect V2 Sensor. In 337 International Conference on Multimedia and Human-Computer Interaction (MHCI). 338

Gebruers, N. et al., 2010. Monitoring of Physical Activity After Stroke: A Systematic Review of 339

Accelerometry-Based Measures. Archives of physical medicine and rehabilitation, 91(2), 340 pp.288–297. 341

Gladstone, D.J., Danells, C.J. & Black, S.E., 2002. The Fugl-Meyer Assessment of Motor 342 Recovery after Stroke: A Critical Review of Its Measurement Properties. 343 Neurorehabilitation and Neural Repair, 16(3), pp.232–240. 344

Gowland, C. et al., 1993. Measuring physical impairment and disability with the Chedoke-345 McMaster Stroke Assessment. Stroke, 24(1), pp.58–63. 346

Guna, J. et al., 2014. An analysis of the precision and reliability of the leap motion sensor and its 347

suitability for static and dynamic tracking. Sensors, 14(2), pp.3702–20. 348

Kamper, D.G. et al., 2002. Alterations in reaching after stroke and their relation to movement 349 direction and impairment severity. Archives of Physical Medicine and Rehabilitation, 83(5), 350 pp.702–707. 351

Khoshelham, K. & Elberink, S.O., 2012. Accuracy and resolution of Kinect depth data for indoor 352

mapping applications. Sensors, 12(2), pp.1437–54. 353

Kitago, T. et al., 2013. Improvement after constraint-induced movement therapy: recovery of 354 normal motor control or task-specific compensation? Neurorehabilitation and neural 355

repair, 27(2), pp.99–109. 356

Kortier, H.G. et al., 2012. Ambulatory assesment of hand kinematics : using an instrumented 357

glove. In XII International Symposium on 3D Analysis of Human Movement. Universita di 358

Bologna, pp. 19–29. 359

Kurillo, G. et al., 2013. Upper extremity reachable workspace evaluation with Kinect. Studies in 360 health technology and informatics, 184, pp.247–53. 361

Lemmens, R.J.M. et al., 2014. Accelerometry measuring the outcome of robot-supported upper 362 limb training in chronic stroke: a randomized controlled trial. PloS one, 9(5), p.e96414. 363

Levin, M.F., 1996. Interjoint coordination during pointing movements is disrupted in spastic 364 hemiparesis. Brain, 119(1), pp.281–293. 365

Milot, M.-H. et al., 2014. Corticospinal Excitability as a Predictor of Functional Gains at the 366 Affected Upper Limb Following Robotic Training in Chronic Stroke Survivors. 367 Neurorehabilitation and neural repair, 28(9), pp.819–827. 368

Mobini, A., Behzadipour, S. & Saadat Foumani, M., 2014. Accuracy of Kinect’s skeleton 369 tracking for upper body rehabilitation applications. Disability and rehabilitation. Assistive 370 technology, 9(4), pp.344–52. 371

Oxford Grice, K. et al., 2003. Adult norms for a commercially available Nine Hole Peg Test for 372

finger dexterity. In American journal of occupational therapy. pp. 570–3. 373

Pastor, I., Hayes, H.A. & Bamberg, S.J.M., 2012. A feasibility study of an upper limb 374 rehabilitation system using Kinect and computer games. Engineering in Medicine and 375 Biology Society (EMBC), 2012 Annual International Conference of the IEEE, 2012, 376 pp.1286–9. 377

Reitan, R.M., 1958. Validity of the Trail Making Test as an indicator of organic brain damage. 378 Perceptual and Motor Skills, 8(3), pp.271–276. 379

Rosenhahn, B., Klette, R. & Metaxas, D. eds., 2008. Human Motion, Dordrecht: Springer 380

Netherlands. 381

Schaefer, S.Y., Haaland, K.Y. & Sainburg, R.L., 2009. Dissociation of initial trajectory and final 382 position errors during visuomotor adaptation following unilateral stroke. Brain research, 383 1298, pp.78–91. 384

Schaefer, S.Y., Haaland, K.Y. & Sainburg, R.L., 2007. Ipsilesional motor deficits following 385

stroke reflect hemispheric specializations for movement control. Brain, 130(8), pp.2146–58. 386

Schieber, M.H. et al., 2009. Selective activation of human finger muscles after stroke or 387 amputation. Advances in experimental medicine and biology, 629, pp.559–75. 388

Scott, S.H., 1999. Apparatus for measuring and perturbing shoulder and elbow joint positions 389 and torques during reaching. Journal of Neuroscience Methods, 89(2), pp.119–127. 390

Simone, L.K. & Kamper, D.G., 2005. Design considerations for a wearable monitor to measure 391 finger posture. Journal of neuroengineering and rehabilitation, 2(1), p.5. 392

Slijper, A. et al., 2014. Computer game-based upper extremity training in the home environment 393 in stroke persons: a single subject design. Journal of neuroengineering and rehabilitation, 394

11(1), p.35. 395

Van der Smagt, P. et al., 2009. Robotics of human movements. Journal of physiology, Paris, 396 103(3-5), pp.119–32. 397

Thewlis, D. et al., 2013. Next-generation low-cost motion capture systems can provide 398

comparable spatial accuracy to high-end systems. Journal of applied biomechanics, 29(1), 399 pp.112–7. 400

Weichert, F. et al., 2013. Analysis of the accuracy and robustness of the leap motion controller. 401

Sensors, 13(5), pp.6380–93. 402

Windolf, M., Götzen, N. & Morlock, M., 2008. Systematic accuracy and precision analysis of 403

video motion capturing systems--exemplified on the Vicon-460 system. Journal of 404 biomechanics, 41(12), pp.2776–80. 405

406

407

Table 1: Descriptive statistics of RMS error, sampling rate, and viable trials by condition (Touch/No Touch) and Target number. Means (SD) are reported where available.

Target

1 2 3 4 5

Touch No

Touch Touch No

Touch Touch No

Touch Touch No

Touch Touch No

Touch

RMS error [mm]

79.5 (45.7)

25.5 (4.1)

23.2 (9.4)

15.4 (3.1)

11.1 (2.7)

14.8 (2.8)

15.3 (4.9)

20.0 (7.2)

58.8 (41.8)

25.0 (9.7)

sampling rate [Hz]

68.1 (25.4)

76.3 (20.1)

52.0 (28.4)

74.3 (13.8)

54.8 (18.9)

76.0 (14.3)

55.1 (23.6)

81.8 (15.7)

35.8 (10.7)

78.2 (15.1)

Viable data [% trials]

4/15 (27%)

27/27 (100%)

11/15 (73%)

26/27 (96%)

11/15 (73%)

27/27 (100%)

13/15 (87%)

27/27 (100%)

8/15 (53%)

24/27 (89%)

6 7 8* 9 10

Touch No

Touch Touch No

Touch Touch No

Touch Touch No

Touch Touch No

Touch

RMS error [mm] -

24.1 (5.1)

17.8 (15.1)

12.1 (5.5)

1.0 (0.3)

4.1 (6.5)

7.1 (2.7)

7.3 (2.4) -

18.8 (10.3)

sampling rate [Hz] -

64.2 (16.0)

55.7 (16.8)

71.2 (19.4)

62.3 (13.0)

63.2 (16.7)

46.0 (13.2)

64.2 (22.2) -

66.9 (20.7)

Viable data [% trials]

0/15 (0%)

25/27 (93%)

14/15 (93%)

26/27 (96%)

15/15 (100%)

24/27 (89%)

12/15 (80%)

21/27 (78%)

0/15 (0%)

21/27 (78%)

11 12 13 14 15

Touch No

Touch Touch No

Touch Touch No

Touch Touch No

Touch Touch No

Touch

RMS error [mm] 249.6 - -

27.4 (11.2)

18.4 (8.5)

18.7 (9.1) - - 266.2 -

sampling rate [Hz] 41.5 73.7 -

32.4 (18.2)

48.0 (28.8)

68.2 (20.8) 61.9 30.7 17.2 -

Viable data [% trials]

1/15 (7%)

1/27 (4%)

0/15 (0%)

11/27 (41%)

8/15 (53%)

12/27 (44%)

1/15 (7%)

1/27 (4%)

1/15 (7%)

0/27 (0%)

* indicates reference target for both systems

408

409

410

411

412

413

414

415

416

417

Table 2: Contingency tables of viable trials

Touching vs. No-Touching

Viable Data No Data % Viable

Touching 99 126 44%

No-Touching 145 80 64%

Fisher’s Exact test, p < 0.001

Outside vs Center targets

Viable Data No Data % Viable

Outside 98 64 60%

Center 63 18 78%

Fisher’s Exact test, p = 0.009

418

Figure Captions 419

Figure 1.Schematic of the experimental design. 420

Figure 2.Correlation between Optotrak and Leap systems for horizontal (left panel) and vertical 421

(right panel) finger position. 422

Figure 3.Bland-Altman plots for horizontal (left panel) and vertical (right panel) for the No-423

Touching condition. 424

425

426

427

428

429

430

Figure 1 431

432

433

434 435

436

437