You are on page 1of 16

Predictions By Means Of

Human Gait Analysis


Rachel Lobl, Tak Maga, and Grant Fehring

STEM Marin
Senior Engineering
September and October Two Thousand and Nineteen
Lobl, Maga, Fehring I 1

Abstract
By using an app that measures gForces in each axis, we made graphs of 21 individual
people’s gait, or walking patterns, with the goal of making regression lines to predict
height or weight solely based on gait. The graphs showed that there is no correlation
between the time it took to take a step and the amount of force the step had.
Additionally, we found no correlation between a person’s height the the force of their
step.

However, we did find a correlation between the time it took to take a step and a
person’s height. Using these conclusions, we are able to build a model to predict
how fast a person’s steps are when normally walking based on their height. This
model logically makes sense since shorter people must take more steps to travel
the same distance as a taller person.

However, this model is not realistically predictive, with high error percentages and a
failure to take other affecting factors into account, such as age. With more data and
an alternative analysis approach, it is possible to obtain a more predictive height
model based on gait.

Table of Contents
Abstract 1
Lobl, Maga, Fehring I 2

Table of Contents 2

Introduction 3
Background and Purpose 3
Fundamental Equations 3
Discovery Question 3

Methods 4
Overview 4
Apparatus 4
Procedures 6

Results 6
Uncertainties 7

Discussion 8
Equations 8
Conclusions 8
Discovery Question Conclusion 10
Anomalies, Error 10
Future Work 11

References 12

Appendices 13
Appendix A - Survey 13
Appendix B- Summative Data 14
Appendix C- Raw gFx Data 14
Appendix D - Raw Individual gFx, gFy, gFz, gFTotal Data 14
Appendix E - Survey Results 15
Lobl, Maga, Fehring I 3

Introduction
Background and Purpose
There are many factors that affect human walking patterns. Through this study, we
hope to make a predictive model that, based on a few of these factors, can predict
aspects of a person’s gait. By studying human walking patterns, we hope to better
understand the factors that enable bipedal motion.

These factors can be used in reverse to predict human characteristics, such as


height. This can have applications in law enforcement, as a person’s gait is a
distinctive marker that is usually not concealed.

Gait analysis also has applications in medical fields, however those are beyond the
predictive aspect of our analysis. This consisted mainly of the force and timing of
each step, to identify mannerisms and general characteristics that can provide
predictive qualities.

Fundamental Equations
We analyzed our subject’s gaits as periodic functions, with:
1
f=
t
With f equaling frequency and t equaling period. λ equals wavelength and c
equaling wave speed.
c
f=
λ
This allows for our analysis of the graphs to be of approximate sinusoidal functions.
See discussion section for application.

Discovery Question

Can we predict, with marginal error, an individual’s height or weight based solely on
their gait?
Lobl, Maga, Fehring I 4

Methods
Overview
Test subjects were given a survey to self-report data concerning their height, weight,
athleticism, gender, and shoe size. They were then fitted with a pouch attached to a
belt that was secured to the torso of the subject. A cellular device running a program
to record force was placed inside the pouch. The subject would then walk a 600 cm
course in a straight line, and stop, whereupon the device would be removed and the
recording stopped. A total of 21 test subjects were tested.

Apparatus
The cellular device used was an iPhone 6s, using its internal sensors and running the
g-Force meter found on Vieyra Software’s Physics Toolbox Sensor Suite. The pouch
used was a Buddy Pouch (Image 1), combined with an elastic Running Buddy Belt to
secure the pouch to the torso. The course was marked out using 3 pieces of blue
masking tape, with two parallel lines 600 cm apart perpendicularly bisected with a
third line connecting the two. See Figure 1. See Table 1.

Image 1 - Buddy Pouch


Lobl, Maga, Fehring I 5

End

Start
600 cm
Figure 1 - Testing Setup

Equipment Uncertainty
Blue Masking tape Not Applicable

Measuring Tape + / - 1 cm

Buddy Belt Not Applicable

Buddy Pouch Not Applicable

iPhone 6s Unknown

Vieyra Physics Toolbox Not Applicable

Table 1 - Equipment Table


Lobl, Maga, Fehring I 6

Procedures
Test subjects are selected from the San Marin STEM Engineering class (3rd Period),
with additional subjects being procured from around the school (including STEM
Physics and Engineering 3rd period) in order to obtain demographic balance. The
subjects were given a predetermined Google Form containing the survey questions
found in Appendix A. The survey obtains the self-reported height, weight, gender,
shoe size, and athleticism. 21 subjects were tested, 10 girls, 9 boys, and two teachers,
one male, one female.

Immediately following the survey, the test administrator would fit the subject with
the buddy pouch using the buddy belt, making sure to place the pouch on the back
of the middle of the torso. The device running the recording app is turned on and
placed in the pouch, which is then sealed using the velcro on the buddy pouch.
Starting from a resting position behind the first parallel line marked on the floor, the
participants walk at a normal pace along the bisecting line to the second parallel
line. Upon reaching the second line, the participants stop and wait for approximately
five seconds. The phone is then removed and recording stopped.

Results
Results are generalized from all subjects, as discussed in the following section. See
Appendix B for the numeric data graphed below. See Appendix C for the complete
data from each participant. See Appendix D for full and complete information.

Graphs
Lobl, Maga, Fehring I 7

1) Time to Take A Step versus Height 2) Height versus Time to Take A Step

3) Weight versus Force of Step 4) Time to Take A Step versus Force in


Each Step

Uncertainties
The force of the step calculations were an average based on three distinct cycles in
each individual's graph. The irregularities in the amplitudes could be caused by the
elasticity of the belt which may have bounced differently in response to a person
taking a step.

There is a possibility of uncertainty in the capturing of data through the Vieyra


software, however uncertainty measurements were not attainable and thus are
assumed to be negligible for our purposes.

Discussion
Equations
Equations are provided by lines of best fit, least squared residual lines. Calculated
Lobl, Maga, Fehring I 8

using Google Sheets.


1) Trendline: y = 0.554x + 103
2) Trendline: y = 0.265x + 73.1
3) Trendline: y = 6.06 + -0.158x + 1.1E-03x^2

Conclusions
Our conclusions were formed based upon individual data producing figures of
general force. Two aspects were analyzed, the first being the number of internal
readings between each apex of force, and the second being the general amplitude.
These readings are located in Appendix B.

Based on the graphs, there is no correlation between the time it took to take a step
and the amount of force the step had. Additionally, we found no correlation between
a person’s height the the force of their step. This is somewhat surprising, as one
might reasonably expect a taller person to have a higher weight, and thus a higher
downward force.

There may be several issues with this reasoning. One, as discussed in the following
section, the subjects selected comprised a small part of the general population and
are not representative of all weights and heights. Second, one’s individual
mannerisms in walking most likely eclipse any effect that weight has on the general
force (gFx in Appendix B).

There is a correlation between the time it took to take a step and a person’s height,
and thusly the inverse is also true. We predict this can be attributed to the larger leg
length that is associated with a larger height. This leads to a longer walk length, and
therefore a longer walk time, measured in cycles of measurement.

Using linear and polynomial regression, we were able to establish a series of


equations, detailed above. These are not predictive of gait, but rather can predict
individual characteristics based on observed walking patterns. Graph 1 illustrates our
main comparison, with step time to height. Combined with Graph 2 and Functions 1
and 2, it is possible to establish our described relationship.

The drop observable in Graph 1 was unexpected, and could be attributable to


several factors. The most likely is that the three subjects were adversely affected by
the social pressure applied by observing their walking patterns. This caused them to
quicken their pace, and walk differently than their standard canter. The effect of
observation could throw off only a small number of participants, producing the data
Lobl, Maga, Fehring I 9

observable above.

A second explanation is that the mannerisms of those outlier subjects were different
than the standard of other subjects, and obscured any effect that height might have
on their walking patterns. This is another plausible outlier explanation.

If more subjects were tested, such outliers would become more obvious, and our
trendline may become more linear positive, such as Graph 2. This would make more
sense subjectively, as physiologically speaking, step time should not decrease as
height increases at any reasonable point.

Using our model to predict weight fails. As we failed to establish a correlation


between force and weight, the model’s predictions based upon gait do not provide
remotely accurate predictions.

In the realm of height, our predictive model fared better. Equation 1 provides our
model for a positive correlation between gait time and height. The model is
reasonably accurate becomes less so the farther away from the mean height or gait.
In essence, all the model tells us is that a person with average gait will probably have
average height, a practically useless conclusion.

In terms of prediction, the model fails to take into account the lessening difference
that occurs with higher and lower heights in terms of gaits, leading to the failure of
the linear model. A polynomial model may provide better results, but the current
model does not provide useful results, due to the unacceptable degree of difference
between estimated and actual results. See Graph 1.

Discovery Question Conclusion


After attempting to fit lines to the data we collected, it seems that predicting height
and weight based on an individual’s gait can be done, but there is certainly no
guarantee that it will be accurate, as our regression line’s correlation is decent, but
not strong enough to be confidently consistent. Therefore, based on the data we
collect, we must answer the question: “Can we predict, with marginal error, an
individual’s height or weight based solely on their gait?” in the negative at this time.

Anomalies, Error
Our data was taken from subjects in their Freshman and Senior year of high school.
Lobl, Maga, Fehring I 10

This leads to a several year age difference in the subjects which could negatively
affect the accuracy of the predictive models. We did not take each test subject’s age
and therefore are unable to account for it when analysing the data.

Error margins are also increased through the use of the buddy pouch. As discussed
earlier, there is a chance that the jostling of the phone while inside the pouch was
great enough to have a significant impact on the results. As seen in certain sections
5, 9, and 12 (among others) of Appendix D, there are some variances in the graphs
that are not explainable by normal walking.

There are also concerns over the collection rate in the Vieyra Software. There are
some positions, as seen in Appendix C, that do not have a corresponding apex force
value. Estimations were made using sinusoidal approximations, but were probably
inaccurate. The difference was probably not great enough to cause any great change
in the results, but that is unknown.

The capacity for a faster collection rate, perhaps on a newer phone or with different
software, may alleviate the problem by providing established data points.

The anomalies found in Graph 1 are discussed previously. The absence of a


correlation in Graphs 3 and 4 are considered findings, and not anomalies, and are
discussed earlier.

Future Work
If we were to complete this project again, we would measure the length of each
participant's step. We believe this would correlate to a person’s height and weight
and would therefore assist us in making a predictive model. We tried to ascertain this
information with our current data but was unable to finish due to time constraints
and the means through which the data was taken.

We hope to apply what we learned about effective surveying in future projects. Our
survey (Appendix A) was ambiguous regarding units and the format we expected
answers in. For example, some subjects answered in feet while others answered in
inches. We were hoping for the data in centimeters, which meant that we had to
convert all of the data by hand.

Our survey also included our hopes to use athleticism to inform decisions about
weight and amplitude of total force (gFx). Due to time constraints, we were unable to
make these considerations.
Lobl, Maga, Fehring I 11

Another archived aspect in our analysis was the differences that footwear makes on
one’s gait. Further tests to determine the effects, if any, shoes have on gait patterns
would be required to make any determination.

Finally, age was not taken into account when analyzing our current data. However, in
the future, it would be pertinent to explore the gait of children compared to adults.
We measured the gait of two teachers but the data was not used in our predictions
because of the age differences and the fear that it might negatively affect the
conclusiveness of the data.
Lobl, Maga, Fehring I 12

References
Debi, Ronen et al. “Differences in gait patterns, pain, function and quality of life
between males and females with knee osteoarthritis: a clinical trial.” BMC
musculoskeletal disorders vol. 10 127. 13 Oct. 2009, doi:10.1186/1471-2474-10-127

Ko, Seung-uk et al. “Sex-specific differences in gait patterns of healthy older adults:
results from the Baltimore Longitudinal Study of Aging.” Journal of biomechanics vol.
44,10 (2011): 1974-9. doi:10.1016/j.jbiomech.2011.05.005
Elbaz, Alexis et al. “The gait speed advantage of taller stature is lost with age.”
Scientific reports vol. 8,1 1485. 24 Jan. 2018, doi:10.1038/s41598-018-19882-1

V. Agostini, A. Nascimbeni, F. Di Nardo, S. Fioretti, L. Burattini and M. Knaflitz,


"Dependence of gait parameters on height in typically developing children," 2015
37th Annual International Conference of the IEEE Engineering in Medicine and
Biology Society (EMBC), Milan, 2015, pp. 7598-7601.
doi: 10.1109/EMBC.2015.7320151
Lobl, Maga, Fehring I 13

Appendices
Appendix A - Survey
Lobl, Maga, Fehring I 14

Appendix B- Summative Data

Amplitude Rows per right cycle


(Force of the step) (the time it took to take a step) Height (cm) Weight (Kg)
Test Subject A 0.655 109 150 56.23582766
Test Subject B 1.1513 111 162.6 46.2585034
Test Subject C 0.7483 112 155 55.32879819
Test Subject D 0.843 113 170 56.6893424
Test Subject E 1.79 113 157.5 46.2585034
Test Subject F 1.163 114 175 54.42176871
Test Subject G 0.9943 114 167.6 54.42176871
Test Subject H 0.776 114 165 50.34013605
Test Subject I 0.8273 115 185.4 77.09750567
Test Subject J 1.113 116 170 62.58503401
Test Subject K 0.9613 116 165 61.2244898
Test Subject L 1.457 116 167.6 79.36507937
Test Subject M 0.7216 120 175 63.49206349
Test Subject N 1.04 121 170 83.90022676
Test Subject O 1.1343 122 180 68.02721088
Test Subject P 0.6753 126 178 61.2244898
Test Subject Q 0.5826 128 165 64.85260771
Test Subject S 1.2353 128.5 178 58.9569161
Test Subject T 0.597 129 160 63.49206349

Appendix C- Raw gFx Data

See attached

Appendix D - Raw Individual gFx, gFy, gFz, gFTotal Data

https://tinyurl.com/rawgaitdata

Appendix E - Survey Results


*Names were given on the original survey but they were excluded to retain the subject’s privacy

Weight
(lb) Weight (kg) Height (cm) Gender Athleticism Shoe Size
Test Subject A 124 56.23582766 150 Female 4 youth two
Lobl, Maga, Fehring I 15

in kids
Test Subject B 102 46.2585034 162.6 Female 8 7
Test Subject C 122 55.32879819 155 Female 8 7
Test Subject D 125 56.6893424 170 Male 8 8.5
Test Subject E 102 46.2585034 157.5 Female 7 61/2
Test Subject F 120 54.42176871 175 Male 5 9
Test Subject G 120 54.42176871 167.6 Male 10 8.5
Test Subject H 111 50.34013605 165 Female 7 8
Test Subject I 170 77.09750567 185.4 Male 7 11
Test Subject J 138 62.58503401 170 Male 8 10.5
Test Subject K 135 61.2244898 165 Female 5 6
Test Subject L 175 79.36507937 167.6 Female 8 11
Test Subject M 140 63.49206349 175 Male 3 9.5
Test Subject N 185 83.90022676 170 Female 9 9.5
Test Subject O 150 68.02721088 180 Male 6 11
Test Subject P 135 61.2244898 178 Male 7 11
Test Subject Q 143 64.85260771 165 Female 1 9.5
Test Subject S 130 58.9569161 178 Male 7 9
Women's
Test Subject T 140 63.49206349 160 Female 4 Size 8

You might also like