PDA

View Full Version : New(?) camera synchronization method



unknown user
08-11-1994, 09:33 AM
Dear subscribers, esteemed collegues and friends,

I need your opinion about an unexpensive and apparently simple
method for synchronizing video data from two cameras.

When two cameras are used to perform a 3D biomechanical analysis
you either need:

a) to synchronize them by means of a GENLOCK device, so that
they are driven at the same speed AND IN PHASE, or
b) to know exactly the speed of each camera, and the (likely
different) time of at least one field for each camera. For example,
the first field recorded by camera 1 (on tape 1) may occur exactly
at 8:00AM, while the first field recorded by camera 2 (on tape 2) may
occur 0.333 seconds later (i.e. 16.65 fields later, with a 50Hz
recording speed). In this case, it is very likely that most fields
recorded on tape 1 will be simultaneous with NONE of the fields on
tape 2. However, interpolation by means of quintic spline can be
later applied to artificially synchronize the digitized video data.

I will call "OFFSET" the 0.333 s difference between the time
of FIELD 0 on TAPE 2 and the time of FIELD 0 on TAPE 1.

Case (a) is simpler, but you need to use a GENLOCK device
for synchronizing the cameras.
In my next project, I will have to place the cameras at a
distance of about 100 m from each other (no, it's not a typo!!!).
I won't be able to connect them by means of a GENLOCK (by the way, we
don't have a genlock, and we have a very limited budget). That's the
reason why I started to experiment simple and unexpensive methods to
measure the OFFSET.

I had an idea that would solve the problem, and I did some
experiments with non-conclusive results. I would be very glad to
know your advice or opinion about it:

While both the cameras are recording, a BEEP (or a series of
beeps) will be produced by two receivers placed in front of the
camera microphones (we do have two walkie-talkies, and a transmitter).
Note that the eventual transmission delay (some technicians warned me
about that) will be present for both receivers.
[incidentally, do you have any information about the possibility that
this delay might be non-negligibly different for the two receivers?]
By means of a ROLAND digital video tape recorder, the audio
signal will be analyzed with a resolution of 1/40 of a frame (each frame
contains two fields, and with a recording frequency of 25 frames/s
the resolution will be equal to 0.001 s).
Thus, the position of the ONSET of the BEEP will be exactly
located on each tape, relative to the video signals, and the OFFSET
between the two tapes determined with extreme accuracy.

[For instance, if the ONSET of the beep will be located at frame
16.34 of camera 1, and at frame 11.50 of camera 2, the OFFSET will be

OFFSET = 16.34 * Hz1 - 11.50 * Hz2

where Hz1 and Hz2 are the frequencies (frames/s) of the two
cameras.]

Is there anyone who ever experimented this method? Can you
find any flaw? Unfortunately, I found a possible source of error in
this method, but my technical knowledge does not suffice to
evaluate it: what if the audio recording heads are placed at a
different distance from the respective video recording heads, in
the two videotape recorders? In fact, I believe that the video recorders
have separate recording heads for AUDIO and VIDEO signals. As far
as I can understand it, the consequence is that the audio signal produced
at a given instant is recorded on the tape some distance BEFORE OR
AFTER the video signal produced at the same instant. This error would be
systematic and constant. This means that it COULD BE MEASURED BY MEANS
OF A LAB TEST AND LATER CORRECTED, but this will be possible only if
no other sources of systematic error are present in the method...

...That's why I need your advice. What do you think about that?
Thanks for your attention,

With kind regards,

Paolo de Leva
Biomechanics Lab.
Istituto Superiore di Educazione Fisica
P. L. de Bosis, 6
00195 ROME, ITALY

Tel.+FAX: (396)-575.40.81

E-MAIL address: DELEVA@RISCcics.ING.UniRoma1.IT

P.S. I assume that each camera will record at a constant speed
(electronically controlled by means of quartz).
By the way, my data shows that the speeds of our two cameras
(battery powered) are almost exactly the same, with an error of
0.0002 fields per second (=0.004%).