View Full Version : answers camera positioning

Renata Kirkwood
03-31-2005, 10:26 AM
Dear All,

I posted a question on about camera positioning. Below are the original
post and the responses I've got. Many thanks to those who replied, I am
sure is going to help us.

1) I have never used Qualisys specifically, but have a lot of experience
with 3D
gait analysis with Vicon (Oxford Metrics) and Optotrak (Norther
Digital), as
well as 3D reconstruction principles in general. I would expect that as
long as your camera system calibration/linearisation procedures have
been done to specifications, the shape of your frontal and transverse
curves would have more to do with marker placement/alignment with
respect to anatomical landmarks, joint axes, etc as per the
biomechanical model. Since frontal and transverse relative joint
kinematics for the lower limbs during gait have smaller dynamic ranges
of motion than sagittal ones, they very sensitive to even slightly
erroneous marker placement. The sagittal angles are also influenced, but
the shape of the curve may still appear 'normal' simply because the
signal-to-noise is better for those larger range measures. Marker
placement/alignment is very specific to the biomechanical model being
used, so make sure to follow the specs of your model. By the way, if
participant has pronounced transverse ("rotational") or frontal
("valgus/varus") deformities, the relative joint angle calculations may
also be distorted... just -how- the 3 angles are portioned out becomes

Regarding the camera positioning question... if you are using 4 cameras
for unilateral gait analysis, I would suggest you start with 1 camera at
each end of the walkway (anterior and posterior views of the
participant, but slightly off axis from one another to avoid
reflections), and the two side cameras so
cameras are at about 60 degree intervals around the one side of the room
(180 degrees). I found better coverage with the front corner camera a
little higher than the others. You can play with camera height, AP
versus ML distances from camera to calibration volume, camera pitch
angle, etc. There is a trade-off between the dimensions of your
calibrated volume (so how many steps you get per walk) and effective
resolution. In addition, false markers due to glare/reflections from
other cameras or other light sources, or noise from other nearby
equpment should be avoided. During the calibration, make sure that all
4 cameras can see all of the calibration markers in your calibration
cluster. Usually, these camera calibration algorithms are based upon
some initial estimate (even from a tape measure) of the x,y,z
coordinates of each of those calibration markers, and of each camera in
a global lab reference system.
Best of luck; I hope some of this info helps, and please let me know how
it turns out.
Stephen W. Hill, Ph.D.

2) Are both legs being collected in the same trail? Are more markers
positioned on the lateral, medial or both sides of the leg? There are a
couple of different set up I would suggest trying. If you are collecting
one leg, position the cameras in a semi-circle shape on the side of the
collection leg. Making sure that at least two of the cameras are
positioned at different heights. If you are collecting both legs, I
would try more of a corner method. Place a camera at each corner of your
capture volume. Once again you can very the height of the cameras. Have
each camera take a different quadrant of the capture volume. This will
help minimize the cameras looking directly at each other. There are
several different ways to tackle this problem. These are just a few
suggestions. In my experience camera position, height and collection
settings all directly affect the final collection.
ope these suggestions help.
Chris Kvamme

3) Here's a few basic suggestions - they may be things you've tried already:
Every marker needs to be seen by at least 2 cameras at all times. If
they can be seen by more cameras, the data will be better. Try arranging
the cameras in an evenly spaced arc around the leg, rather than having
cameras directly in front and behind.- If you use hemispherical, rather
than flat, markers, the cameras will see them much more easily, even
when the leg twists away from them. We make these cheaply by cutting
polystyrene balls (available in various sizes from craft shops) in half
and covering with reflective tape. - The depth resolution will probably
be better if the cameras have fairly different views of the markers. Try
having two cameras lower than the other two, and two cameras a bit
further away than the other two (so that each camera has a different
height/distance combination).
Good luck!
Sharon R. Bullimore

4) Hi there, I am Jebb, and I am a student at UMass Amherst. We have two
Qualisys camera systems, and I alos work for innovision systems, the
notrth american distributerfor Qualisys. As to your set-up, the optimal
camera angle is 60 degrees between each to get the best depth
resoloution (away from the cameras). This information is in the appendix
of the manual. So use a protractor, and stand in the center of your
collection area. Try to optimize the apparent angle (to near 60) from
the lens of one camera to the center of the collection to the lens of
another camera, by repositioning the cameras. A square configuration
sounds reasonable, either one high one low one anterior one posterior,
or two high two low, with a good seperation between. I can talk more if
you need...
Jebb Remelius

5) You want to position the cameras at a 45 degree angle to the line of
\ | /
/ | \

This would help with coverage, it also depends on your marker set (whether
they are all lateral or a combination of anterior and lateral). A 5th
camera would be optimal!

Alan Morris, MASc., PEng.

6) I've had similar problems using 3- and 4-camera configurations. What
helped was staggering the heights of the cameras so they "see" the
markers from different angles. Also, make sure the two cameras on the
test limb side are at different angles, not parallel to each other,
since they will essentially be seeing the same thing, rather than giving
two different perspectives. Finally, somethng the salesmen won't tell
you, is that the top labs use 6- and 8-camera systems for a
reason...they work better!

Good luck.
Jim Phillips
Seton Hall University

7) I have been using a QTM motion analysis system at Rush Presbyterian
St. Luke's Medical Center in Chicago. We had 4 ProReflex 240 MCU's.
Camera positioning may perhaps not be the only thing responsible for
your troubles. I don't know what MCU's you are using, but have you tried
focusing the cameras? The cameras have two rings of LED's which
correspond to focusing on objects near and far. There is a sliding door
on top of each camera which gives access to the focus aperture.

Adjust the camera tripods so that their height is staggered ( not all on
the same level ) and they are focused right on top of the force plate.
In my opinion the cameras are not very good, and loose markers very
rapidly as soon as the subject moves to the edge of the capture volume.
Best of luck !
Amardeep Singh Dugal

8) I cannot speak directly for the Qualisys system, but perhaps for
motion capture setups more generally. One suggested metric for camera
placement is 2.5 times the subject's height (or point of interest). This
is principally a function of the camera's viewing angle, but must also
account for the distance from the subject, as camera resolution is
likely essential for your analysis. Another critical aspect is that you
want at least THREE cameras to see EACH individual marker, to
reconstruct the 3D trajectories. If you position cameras directly in
front and in back (as you suggested), then only one camera is going to
"see" a marker placed on that side of the body. If the cameras are
positioned in oblique orientations, camera redundancy (a GOOD thing) is

I've attached a picture from UGS' Jack, that show one possible camera
setup for a one-sided gait analysis. Ideally you want MORE cameras to
reduce marker occlusion or dropout, but use what you have and be
intentional about what you intend to capture. Decide specifically 1)
which data you need, 2) which data you might need, 3) which data you
want, 4) which data you might want, and 5) anything else. Position
cameras according to this hierarchy.
Good luck.
Kevin Rider

9) I have not used Visual 3D. We did collect about 20 subjects with the
Qualisys system, and the data ended up being ok. Our frontal and
transverse plane data were ok. However, I don't really trust these
angles measure using any system - we can talk about that later perhaps.
Our main issue with the Qualisys had to do with the amount of manual
intervention required in tracking the markers. It was very labour
intensive, and we lost quite a few trials. Once we had meticulously
tracked all of the markers (and we were using 16 for gait) the data was
good. We had a 5 markers on the thigh (one on the greater trochanter, 3
on a plastic disc that was strapped to the midthigh, and one on the
lateral epicondyle. On the shank we had another 5 (one on the fibula
head, 3 on a plastic disc strapped to the mid shank and one on the
lateral maleoli). I think we ended up removing the one on the head of
the fibula because it was too close to the lateral epicondyle. (I can
check for sure if you need to know). Now, it took us a long time to be
able to get good data. I think we worked for months and months to get a
camera set-up and calibration protocol that worked for us. This is
critical for good data. We had best results when the cameras were close
to the ground and arranged in a semicircle about the test area. We were
using 5 cameras, and were careful that none of the cameras were pointed
at each other. Calibration was hard. We used the wand calibration, and
there was only one of my students who was able to perform the
calibration with satisfactory results.
Kevin J. Deluzio

10) In our experience, a marker must be 'seen' continuously by at least
three cameras for optimal data collection. From your description it
seems that your markers are on the lateral side of one leg. Maybe you
can try to position your cameras as follows: one at 30° with respect to
the walking direction, one at 70°, one at 110° and the last one at 150°
(thus no camera view parallel to the walking direction). I 'm not sure
whether this will help, but we also ended up with a successful
positioning of the cameras after a lot of trial and error...
good luck
Matthieu Lenoir

11) We currently use a 5 camera ProReflex system along with Visual 3D,
however, we have just got another 2 cameras. in my experience 5 is an
absolute minimum if you are using CAST (clusters) we generally have an
arc of 3 at the from of the subject with 2 positioned anteriorly. My
best advice to you would be to find the money somewhere to get at least
another camera. I have to say that 5 cameras has provided us with very
good data and that your choice of Visual 3d was very wise as it is an
incredible package.
Dominic Thewlis

12) We have also got a new Qualisys system and have played with camera
positioning (we've tried it with 8 cameras). My suggestion is not to
place the cameras exclusively perpendicular to the different planes of
motion. Rather than placing the cameras on the four sides of a square,
try the four corners of a square.
I'd be interested in hearing what the others respond.
Good luck.

Renata Kirkwood, Ph.D.
Universidade Federal de Minas Gerais
Departamento de Fisioterapia
Belo Horizonte - Brasil
Tel: 31 3499-4782
Cel: 31 9985-0707
email: renatak@forusers.com