View Full Version : wide angle lens correction and joining two camera views

Marguerite Butler
04-02-2003, 04:51 AM
Dear Biomechanists,

I am new to video analysis of locomotion and hoping that someone can
point me in the right direction. I am forced to use wide angle lenses
and am having a hard time correcting the errors.

I am filming iguanas (which are large lizards) running along a trackway
to get data related to escape behaviors (burst speed, acceleration,
time to constant average speed, stride length, etc.). I have determined
that I need to capture the first 1.8M of their run, and because my room
is too small, I cannot move my cameras back farther than about 8ft. I
am using two cameras (cam 1 captures the first 105cm of the run, cam 2
captures the 105cm, with a region of overlap of about 20cm). I am using
8mm lenses. I am able to get good video, but it is very hard to get the
tracked points to match up between the two cameras.

I am currently filming from above, so I get a dorsal view of the
animal. Big errors creep in when the animal deviates from the plane in
which the calibration was taken (the distance from the animal to the
camera changes). I have been able to figure out a
"back-of-the-envelope" correction using simple trigonometry (assuming
the camera is perpendicular to the track and assuming or estimating the
height of the animal), but I was wondering if there was a more rigorous
approach or discussion of the problem. I have looked through the
archives and also through some biomechanics textbooks, but it seems
that most people just try to avoid this problem rather than finding a
way to work with it.

Here is an ascii diagram of my solution, assuming that X is horizontal
and Y is vertical:
The camera is above the animal, perpendicular to the track, and the
calibration marks are on the track floor.

| \
| \
| \
| \
| X' animal's head moving along track -->
| \
---------- X track floor; apparent location of animal's head as viewed
by camera
^^^ error in estimating animal's position along
direction of motion

Y = perpendicular distance of camera to track floor
X = Apparent distance of animal from center of camera view.
Y' = height of animal
X' = error (difference between animal's actual location and apparent

I used the relation tan(theta) = Y/X = Y'/X' where theta is the angle
between the camera and the track at the location of the animal.

If I guess at a value for Y' and measure Y and X, I can solve for X'. I
don't have money to buy a third high-speed video camera, but I'm trying
to record a lateral view with a DV camcorder to at least get a better
estimate of the height of the animal (or the tracked points).

Is there a better way? Can I use the information from the overlap
region of the second camera?

I know that I can get a better match between the cameras if the
calibration points are at the height of the markers (rather than on the
floor), but the animal does not remain perfectly at the same height
throughout the run (especially the legs), and different iguanas will
vary in height off the floor.

I have tried filming lateral views, but this was even more problematic
because the animal's run has greater potential for deviation to the
side of the track and also because I have to film through plexiglass.
I'm using reflective markers so the glare from the lights was really
hard to eliminate.

If anyone can help with suggestions or references for me to look into,
I'd greatly appreciate it!

Thanks for you time,
Marguerite Butler
__________________________________________________ ______________________

Marguerite A. Butler
Department of Ecology and Evolutionary Biology
University of Tennessee
569 Dabney Hall
Knoxville, TN 37996-1610

Phone: 865-974-7894
FAX: 865-974-3067

To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l