Following is the original message:

************************************************** *****************
Dear Biomech-L readers,

I am a MPhil student. I am now going to compute a pascal program for stride
length calculation in running or gait analysis. The purpose of this program
was to reduce the time consuming during motion analysis data processing.
Based on calculating the stride length, the instant of stride touch down
(film number) must obtained first. However, it was found that the instant of
heel touch down or toe takeoff were difficult to calculate throught programming.

I know that some gait analysis system synchronized with video capture and
force platform, so the heel touch down and toe take off were easy to obtain.
Some commercial motion analysis softwares were need to enter the instant
(film number) of touch down or take off by human observation when
calculating of stride length without force platform.

Now the problem is how to find out the instant of heel touch down or toe
take off with only providing the raw data of the heel/toe x,y-coordination.
The local minimum point of the heel/toe y-coordination is not a good way to
obtain the touch down or take off instant, since sport shoes or running
shoes contains cushions. The minimum point of heel/toe may not represnet the
touch down or take off instant respectively. Is there any suggestions about
how to determinate the instant of heel touch down or toe take off
systemically rather than by human observation ? Any idea or suggestions
would be greatly appreciated. Thank you.

Best regards,

************************************************** *****************

###################### Summary of response: #######################

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~
Foot switches or multiple 2D views of the event to establish the
3D coordinate.

Motion Analysis Corporation
Daniel India, Vice President
3617 Westwind Blvd
Santa Rosa, CA 95403 USA
HQ Tel: 707-579-6500 Direct 847-945-1411
HQ Fax 707-526-0629 Direct 847-945-1442
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~
Hello Jim,

>From your letter to Biomch-L on 29AUG98 it is not clear to me whether you just
need to calculate stride length, or also need to determine the instants of
initial-contact ("heel touch down") and final-contact ("toe take off.") May I
offer the following thoughts, based on many years of thinking about the same
questions at Shriners Hospitals for Children in San Francisco with my late
colleague Roger St. Helen. Roger and I tried a number of methods but never
completed a practical program to automatically identify foot contact events in
the gait cycle.

Background assumptions:
Human gait, either walking or running, is a cyclic process in which similar
movements are repeated over and over again. Gait analysis seeks to better
understand walking or running by making measurements, during several gait
cycles, of variables such as body-segment movements, ground reaction forces,
or the timing of muscle contractions. The gait is then described in terms of
these measurements taken from a representative cycle, or averages of several
cycles, in order to compare the measured gait with "normal" or with the same
subject at a different time.

Temporal/Distance parameters:
The distance moved during each cycle and the time taken for each cycle are the
most fundamental measurements of gait, usually expressed as step or stride
length, cadence (steps per minute or steps per second,) and average velocity
(meters per minute or cm per second.)

To measure these distances and time durations, some event in each gait cycle
must be defined that identifies when one cycle ends and the next cycle begins.
The most commonly used event is initial-contact of the foot with the floor,
but any other repeatable event could be used instead. For example, mid-swing
is an easier event to identify from motion measurements than initial-contact
or final-contact because mid-swing is a high-velocity event and the foot
contact events are low-velocity events.

Swing-Phase and Stance-Phase:
To make the familiar gait-analysis graphs of variables such as joint movements
or the timing of muscle contractions, it is necessary to identify initial-
contact and final-contact so these events can be marked on the graphs to show
which part of the graph is stance-phase and which part is swing-phase.
However, if only the numeric values of step length or the other gait-cycle
temporal/distance parameters (cadence and average velocity) are required, then
the mid-swing event is easier to use for automatic processing by a computer

For example, it is easy to identify the frame where the swinging leg passes
the stance leg by comparing the position of the swinging toe marker with the
position of the stance toe marker. The film or video frame where the swinging
toe marker first passes the stance toe marker, in the direction of
progression, can be defined as "mid-swing," (or "mid-stance.") The distance
between stance toe markers for successive mid-swing frames would then be the
distance between footprints, which probably is the most intuitive definition
of step length.

If the goals of the program you are writing also include identifying initial
and final foot contact, and not just stride length, I offer the following
ideas for your consideration.

Roger and I concluded that there are three different foot-contact sequences
that must be considered to have a useful program: 1. Heel touches the floor
before toe, 2. Toe touches before heel, 3. Heel and toe touch at the same

First of all, the only parts of the gait cycle we are interested in, when
searching for foot-contact events, is when the progressional velocities of
heel and toe markers are low, because we are looking for periods in the gait
cycle when one or the other foot has stopped. In a program, we could specify
that the progressional distance between the positions of a marker in two
successive frames must be less than some small threshold value.

1. Heel touches floor before toe: The toe marker is higher than the heel
marker by some small threshold value when the above specified low velocity is
achieved. In a heel-toe gait, when the heel first touches the floor, the
contact force on the heel suddenly forces the toe down. On a side view plot
of the paths of the heel and toe markers, one can see this sudden change in
direction of the toe marker quite clearly. In a program, one could calculate,
frame by frame, the difference in elevation between the heel and toe markers
and identify the frame where this sudden change occurs.

2. Toe touches floor before heel: The toe marker is lower than the heel
marker by some threshold value when the above specified low velocity is
achieved. In the toe-first contact pattern, the heel marker usually drops and
moves backward as the limb is loaded. This motion could perhaps be detected,
or an adequate condition for contact might be a progressional velocity of the
toe marker that is less than a specified threshold value.

3. Neither of the above: Heel and toe are at about the same elevation as the
foot slows down. I say "about the same" because the threshold values in the
above tests must be bigger than zero to work reliably. In this pattern, the
program could test for progressional velocity less than a specified threshold

It may be that a successful program could be written that used the same toe-
marker velocity test for both conditions 2 and 3, in which case only two
different contact patterns would have to be detected. Note that these tests
assume that markers are put on the foot so that when the foot is flat on the
floor, heel and toe markers are both at the same elevation.

To determine the best sizes for the various threshold values in the above
tests, it will be very desirable to compare the events chosen by the program
with events indicated by simultaneous force platform measurements. Once you
have adjusted the various thresholds and are satisfied that your program is
accurate in relation to the force platform, you can use the program for steps
that do not touch the force platform. For a useful program, I think it may be
more important to get consistent results than to get results that exactly
match the force platform.

Final-contact is harder to define precisely in time than initial-contact
because it happens slower. Force platform comparisons are needed to choose a
progressional velocity threshold for the toe marker beyond which the foot will
be considered to have left the floor. The toe marker definitely starts to
move before the toe leaves the floor, and the force platform is the best way
to determine how much before.

I hope these ideas are useful to you. If you can write a reliable program, it
will be very useful. I would be delighted to hear of your results.

Yours with best wishes for success,

Larry Lamoreux, Ph.D.
Gait Dimension
P.O.Box 1186
Lafayette, CA 94549-1186
Phone: +925-283-7718
FAX: +925-283-5919
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~
Dear Jim,

I have found a simple angle-based method to automatically
detect the Heel-Contact instants only from the trajectories of
a set of markers giving the hip, knee and ankle angles.

The following two abstracts are available:

- CORDIER E.: "Automatic method for cycle extraction and
segmentation in human gait kinematic data", presented in the
XVIth Congress of the International Society of Biomechanics
(ISB'97), Tokyo, 1997.

- CORDIER E.: "Automatic methods for cycle extraction and
segmentation in human slope walking kinematic data",
presented in the 13th International Symposium on Multisensory
Control of Posture and Gait (ISPG'97), Paris, 1997.

I will send you a re-print of the two abstracts as soon
as possible.


E. Cordier

__________________________________________________ _____________

(Dr.) Emmanuel CORDIER - Head of the 'Foot & Ankle Group'

Institut fuer Experimentelle Chirurgie
der Universitaet Ulm

Abteilung Unfallchirurgische Forschung
und Biomechanik

Helmholtzstrasse 14

Phone : direct +49 (0)731 . 50 . 22908
std. +49 (0)731 . 50 . 23481

Fax : +49 (0)731 . 50 . 23498

Email :


__________________________________________________ _____________
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~
Dear Jim,

The topic of stride lenght estimation has been treated in:

Lanshammar H
Estimation of gait cycle duration and stride length from one landmark
kinematic data. 5th Meeting of the European Society of Biomechanics,
Berlin (West), September 1986. In "Biomechanics: Basic and Applied
Research", ed: G Bergmann, R Kölbel, A Rohlmann, Martinus Nijhoff
Publishers, pp 129-134, 1987.

Rather than trying to estimate events like heel contact, the method is
based on the autocorrelation function. During walking the vertical
position of markers attached to a leg will be close to a periodic
function. By minimizing a criterion function defined as the squared sum
of differences between such a time series and a time shifted version of
the same time series it is straightforward to estimate the gait cycle
time. The minimization is done with respect to the time shift.
Obviously, the criterion function has a minimum when the time shift is
zero. This minimum must be excluded. When the time shift increases, the
next minimum will occur when the time shift is equal to the stride cycle

If you have difficulties in finding the paper, let me know, and I will
send you a copy!

Good luck! - Håkan
************************************************** *********************
* Håkan Lanshammar Systems and Control Group, Uppsala University *
* P.O. Box 27, S-751 03 Uppsala, SWEDEN *
* E-mail:, Tel: +46-18-471 30 33, Fax: +46-18-50 36 11*
* WWW: *
************************************************** *********************
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~
Dear Jim,

we use the horizontal velocity (running direction)to find the begin
and end of a stride. With this information we calculate stride
length, duration of stance and duration of motion cycle.

Good luck

Christian Peham

************************************************** ******
* Dr.Christian Peham
* email:
* Clinic for Orthopaedics in Ungulates
* Locomotion Research Group
* University of Veterinary Medicine Vienna
* Phone: +43-1-250 77/5506; Fax: +43-1-250 77/5590
* Josef Baumanngasse 1; A-1210 Wien
************************************************** ******
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~

I have some suggestion for you. In my dissertation (University of
Michigan), I caculated
stride length based on motion data (forward and vertical components) of
infrared diodes. The
markers located at ankles.

Without force plate, it's hard to obtain the real heel touch. But, it's
reasonable to
calculate stride length, since a stride is not necessary defined by heel
touch. I first
filtered my data and differentiated it to find vertical and forward
velocities. Heel touch
would lead to both velocities approaching 0. "0" is a theoretical number.
Numerically, you
may set to be less than 0.1 m/s. My analyses showed that this method is
quite repeatible.

good luck.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~
Hello Jim, You are right - there is only one way to define
toe-off and heelstrike properly, and that is by using force platform
data. (Even here you must be careful to find the y force maxima
carefully.) However if you are able to film people wearing their sports shoes
while they walk across a force platform, and synchronise the film
with the force platform data, you may then find what the position of
the sports shoes looks like during heelstrike and toe-off. Hence you
may then apply this information away from the force platform, safe in
the knowledge that you are correctly identifying these instants.

Mark W Swanepoel, PhD
School of Mechanical Engineering
University of the Witwatersrand
South Africa
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~
I am assuming that when you mention 'x,y-coordinates', y is vertical
and x is horizontal.

You could try using the x-coordinate of the motion--for example,
heel-strike would occur at the instant when the forward motion of the
heel stopped (or, if walking on a treadmill, when the heel marker
reversed direction).

Also, does simply using the y-coordinate of the motion introduce such
a great error even if the subject is wearing padded sneakers, as
compared to simply eyeballing the result while watching video
frame-by-frame (which you are trying to avoid)? I'd be curious to
know how much error is introduced.

Ian Kremenic, M.Eng.
Research Assistant/Network Admin.
Nicholas Institute for Sports Medicine and Athletic Trauma
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~
Have you tried foot switches to register the pressure at foot strike and
toe-off? I know this isn't your approach, but they may be easier than you
motion-based data collection.

Good luck!

Jon Fewster
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~

Thank you very much for all response !

Best regards,


Biomechanics Laboratory
Department of Sports Science and Physical Education
The Chinese University of Hong Kong
Hong Kong
Tel : (852) 2609 6082
Fax : (852) 2603 5781

To unsubscribe send UNSUBSCRIBE BIOMCH-L to
For information and archives: