^{*}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/.)

Wearable sensors for gait analysis are attracting wide interest. In this paper, a wearable ground reaction force (GRF) sensor system and its application to measure extrinsic gait variability are presented. To validate the GRF and centre of pressure (CoP) measurements of the sensor system and examine the effectiveness of the proposed method for gait analysis, we conducted an experimental study on seven volunteer subjects. Based on the assessment of the influence of the sensor system on natural gait, we found that no significant differences were found for almost all measured gait parameters (p-values < 0.05). As for measurement accuracy, the root mean square (RMS) differences for the two transverse components and the vertical component of the GRF were 7.2% ± 0.8% and 9.0% ± 1% of the maximum of each transverse component and 1.5% ± 0.9% of the maximum vertical component of GRF, respectively. The RMS distance between both CoP measurements was 1.4% ± 0.2% of the length of the shoe. The area of CoP distribution on the foot-plate and the average coefficient of variation of the triaxial GRF, are the introduced parameters for analysing extrinsic gait variability. Based on a statistical analysis of the results of the tests with subjects wearing the sensor system, we found that the proposed parameters changed according to walking speed and turning (p-values < 0.05).

The quantitative analysis of gait variability using kinematics and kinetic characterisations can be helpful to medical doctors in monitoring patients’ recovery status in clinical applications. Moreover, these quantitative results may help to strengthen their confidence in the rehabilitation. Walking speed, stride length, the centre of mass (CoM) and the centre of pressure (CoP) have been considered as factors in the evaluation of walking gait [

Many kinds of stationary systems such as force plates and instrumented treadmill devices are available to measure CoP and triaxial GRF [

To overcome such limitations of stationary devices in GRF measurement, many researchers are developing wearable sensors attached to shoes [

In this paper, we describe a new wearable GRF sensor system which has a thin and light sole. Verification of the system’s measurements and evaluation of effect of the system to natural gait are presented. Moreover, we applied the multi-step data of CoP and triaxial GRF obtained from the sensor system to analyze extrinsic gait variability.

As shown in _{xi}, F_{yi}_{zi}_{i}, y_{i}_{5}, y_{5}

In this study, the mid-stance phase (foot-flat) during which the heel and forefoot are in contact with the ground, was evaluated to analyze extrinsic gait variability. All three components of GRF and CoP coordinates (_{o}, y_{o}, z_{o}

A multi-channel data-logger was specially designed using a micro-computer (PIC 16F877A) system for the sensor system, and the data from the sensor system could be saved in an SRAM at a sampling rate of 100 Hz. The proposed algorithm for evaluating extrinsic gait variability was made using MATLAB software (The Mathworks, Natick, MA, USA). The off-line data analysis was performed by uploading data saved in the SRAM to a personal computer through a RS232 communication module. The small triaxial force sensors integrated in the shoe and the portable data-logger based on a micro-computer system were all low energy consumption devices, so the wearable sensor system could be powered using a 300 mAh (NiMH 30R7H) battery.

A combination system composed of a force plate EFP-S-2KNSA12 (KYOWA, Japan) and an optical motion analysis system Hi-DCam (NAC Image Tech., Japan) was used as a reference measurement system to validate the results of the developed sensor system. A low-pass filter with a cut-off frequency of 10 Hz was applied to the measurements of the reference system. By measuring orientations of the global coordinate system fixed on the developed sensor system using the optical motion analysis system, we transformed the stationary force plate measurements in their coordinate system to the global coordinate system of the sensor system, and compared the measurements of the developed sensor system with the reference system using a statistical method. Seven young volunteers (four men and three women: age = 28.5 ± 3.5 years, height = 168.5 ± 5.5 cm, weight = 63.4 ± 9.3 kg) were required to wear the sensor system to walk on the force plate, and the signals from the sensor system and the reference system were simultaneously sampled at 100 samples/s.

In order to apply the developed sensor system to extrinsic gait variability analysis, we designed a test experiment for the application of the sensor system by considering the effects of walking speed and turning on walking gait [

Gait evaluation based on analysis of temporal features of GRF and CoP has commonly been discussed on the basis of measurements of force plates fixed on the ground [_{x}(t)_{y}(t)_{z}(t)_{on}_{off}], and on/off subscripts mark the beginning/end of the foot-flat phase of gait.

To analyze all the successive foot-flat phases _{z1}(t)_{z3}(t))_{z5}(t)

The first condition is to find the periods when _{z1}(t)_{z3}(t))_{z5}(t)_{z1}(t)_{z3}(t))_{z5}(t)

Based on the two selection conditions above, the time sample matrix ^{k}(t)^{k}_{on}^{k}_{off}

The general scheme of the evaluation method consists of two steps:

CoP envelope curves calculated from CoP trajectories of each step during successive walking (if all the CoP trajectories are obtained in a successive trial);

Measures of dispersion of triaxial GRF in the boundary area of the CoP envelope curves.

Envelope curves were used to simplify the gait evaluation algorithm by giving boundary conditions for integration calculation. In the first step (the calculation of the CoP envelope curves), two envelope curves including medial boundary function ^{Medial}: (x^{M}, y^{M})^{Lateral}: (x^{L}, y^{L})^{k}(t)_{posterior}_{anterior}^{M}_{posterior}_{anterior}^{L}_{posterior}_{anterior}

The second step is based on the results obtained from CoP envelope curves estimated in the first step. We adopted the area (_{cop}^{M} (x^{M}, y^{M})^{L}^{L}, y^{L})^{X,Y,Z} : [ACV^{X}, ACV^{Y}, ACV^{Z} ] for three-directional GRF to evaluate walking extrinsic gait variability.

Before validating the accuracy of the sensor system, we used the parameters including stride length (SL), stride width (SW), maximum lateral foot excursion (ME), single stance time (SST), double stance time (DST), stride time (ST), maximum GRF (MaxF) and minimum GRF (MinF) as proposed by Liedtke [

Statistical analyses were performed to determine the effects of walking direction and speed on extrinsic gait variability, which are presented as mean ± standard deviation (SD). Linear mixed effects models [_{cop}^{X}, ACV^{Y} and ACV^{Z} by walking condition (speed and direction) were statistically significant (p < 0.05). The fixed effects for this study were the average differences in _{cop}^{X}, ACV^{Y} and ACV^{Z} for each walking condition using a repeated measures analysis of variance (ANOVA).

As shown in

The RMS distance between both CoP measurements was 2.1 ± 0.4 mm, corresponding to 1.4% ± 0.2% of the length of the shoe. In order to examine whether these reported differences are statistically significant, we quantified the differences between the two measurement systems using the ASTM standard which reports differences in terms of bias and precision. The terms repeatability limit and reproducibility limit are used as specified in Practice E 177 of the ASTM standard [

The 95% repeatability limit (within subject) and 95% reproducibility limit (between subjects) of the triaxial GRF measurements are 2.5 N and 2.5 N for

The parameters used to assess the effect of the sensor system on gait were averaged over ten trials per subject. An overview is presented in

Using the developed algorithm, we extracted GRF data for foot-flat gait phase of each step from the measured spatio-temporal gait data. The CoP envelope curves were calculated from the CoP trajectories of the extracted intervals. The size of the spreading area of the trajectories of CoP on the foot-plate of different steps for the walking with obstacles was greater than that for the walking without obstacles, at all walking speeds. Moreover, the area between the calculated CoP envelope curves increased with accelerating movements (see

The mean and SD of Acop and ACV for walking without obstacles compared to walking with obstacles at three walking speeds (slow, average and fast) for the seven subjects in the three repeated trials are shown in

The force measurements by the sensor system in the

Significant differences between instrumented and normal shoes have been found in the maximum ground reaction force [

As a research application, some sensor systems have been used to evaluate the effects of walking speed and walking direction change on extrinsic gait variability. Kinematic gait parameters calculated using joint motion trajectory have been widely adopted to analyze gait variability [

The effects of walking speed and turning upon gait must be considered when interpreting the effectiveness of these new definitions for extrinsic gait variability. Turning or walking direction changes affect lower limb kinematics, kinetics, and step length [

Walking speed affects kinematics, double-support time, step width and other clinical correlates of stable walking [^{X} and ACV^{Y} augment when subjects increase walking speed, so the average coefficients of variation (ACV) of the ^{Z} obtained from the vertical component of GRF could not clearly reflect the walking speed changes, so they cannot be used to assess speed effects on extrinsic gait variability during level walking.

A wearable GRF sensor system to measure CoP and triaxial GRF in a number of walking trials was developed. Natural gait was almost never affected by the sensor system in the ambulatory GRF measurements. The 95% repeatability limit (within subject) and 95% reproducibility limit (between subjects) of the triaxial GRF measurements is 2.5 N and 2.5 N for ^{X}, ACV^{Y} and ACV^{Z} in walking gait analysis, we implemented an experimental study of a group of healthy subjects who were required to walk under the designed experimental protocol. According to our statistical analysis, the proposed parameters Acop, ACV^{X}, and ACV^{Y} increased with the increase of walking speed (p-values < 0.05), but no similar trend was found for ACV^{Z} with the speed changes. All the parameters of walking with obstacles were greater than walking without obstacles for all three speeds, and these comparisons were statistically significant (see

Although good results were obtained for validation and extrinsic gait variability analysis, the sample size for the application was limited (seven healthy subjects), and the variability was assessed on the aspect of obstacle-avoidance. In future research, experiments on more diverse subjects and various walking conditions will be necessary to support clinical applications of the sensor system.

We thank Ikemoto T. MD and Nakao S. PT from Kochi Medical School for their technical support and all the subjects who participated in this project.

A wearable GRF sensor system constructed using five small triaxial force sensors.

Two sets of trials designed in the experimental protocol. The distance between starting point and destination is 10 m for the two trial paths.

Triaxial GRF measured by the wearable sensor system (solid line) and referenced system (dashed line) during walking trial.

Centre of pressure (CoP) measured by the wearable sensor system (solid line) and reference sensor system (dashed line) referred to a global coordinate system. and referenced sensor system (dashed line) during walking trial.

Representative plots of CoP envelope curves (thick lines) and CoP trajectories for each step in the successive walking trials. CoP trajectories of the six trials were given in the six sub-plots in which the horizontal axis and vertical axis were defined as medial-lateral direction and anterior-posterior direction respectively. Moreover, the left and right envelope curves were plotted using a widening solid line for each trial.

Statistical comparison of the evaluation parameters’ (Acop, ACV^{X}, ACV^{Y} and ACV^{Z}) mean and standard deviations (SD) for all the subjects.

Statistical comparison of evaluation parameters’ (Acop, ACVX, ACVY and ACVZ) mean and standard deviations (SD) of each subject for the three selected speeds required in the two sets of trials. The square markers and deviation scopes represent the evaluation parameters’ mean and standard deviations (SD) in the first set of trials where subjects were required to walk along a lead line; and the diamond markers and deviation scopes are used to represent the second set of trials where subjects were asked to walk bypassing some obstacles.

The mean and standard deviation of the gait parameters including stride length (SL), stride width (SW), maximum lateral foot excursion (ME), single stance time (SST), double stance time (DST), stride time (ST), maximum GRF (MaxF) and minimum GRF (MinF) for all subjects (p-values < 0.05).

| |||||||||
---|---|---|---|---|---|---|---|---|---|

1,440.9 ± 2.0 | 83.8 ± 1.7 | 25.3 ± 1.1 | 0.83 ± 0.08 | 0.16 ± 0.04 | 1.20 ± 0.09 | 709.5 ± 1.3 | 540.9 ± 1.9 | ||

1,427.3 ± 1.3 | 90.2 ± 1.0 | 27.1 ± 0.5 | 0.88 ± 0.03 | 0.18 ± 0.06 | 1.11 ± 0.05 | 689.1 ± 2.2 | 533.7 ± 1.2 |

The four parameters (mean ± SD) for the three walking speeds (slow, preferred and fast) at two sets of walking trials: without obtacles and with obstacles with corresponding p-values. Calculated evaluation parameters including CoP distribution area (_{cop}^{2} and the average coefficient of variation (^{X}, ACV^{Y}^{Z}

Without obtacles | With obstacles | p-value | ||||
---|---|---|---|---|---|---|

_{cop}^{3} |
Slow | 2.9 ± 0.9 | 3.9 ± 1.3 | 0.001 | ||

Preferred | 2.5 ± 0.7 | 4.0 ± 1.3 | 0.007 | |||

Fast | 2.9 ± 0.5 | 4.4 ± 1.5 | 0.021 | |||

^{X} |
Slow | 2.6 ± 0.8 | 2.9 ± 1.3 | 0.013 | ||

Preferred | 3.0 ± 0.8 | 3.5 ± 1.2 | 0.009 | |||

Fast | 3.9 ± 0.8 | 4.3 ± 1.5 | 0.018 | |||

^{Y} |
Slow | 1.4 ± 0.3 | 2.6 ± 0.7 | 0.011 | ||

Preferred | 1.5 ± 0.3 | 3.2 ± 1.0 | 0.019 | |||

Fast | 1.6 ± 0.2 | 4.1 ± 1.6 | 0.023 | |||

^{Z} |
Slow | 11.5 ± 2.1 | 13.9 ± 1.2 | 0.032 | ||

Preferred | 10.2 ± 1.6 | 12.8 ± 2.6 | 0.045 | |||

Fast | 10.3 ± 2.9 | 13.2 ± 3.4 | 0.034 |