PDA

View Full Version : SUMMARY: Videorecording speed variability



unknown user
12-24-1994, 06:38 AM
Thanks to all the subscribers who responded to my plea about
Videorecording speed variability.

This is a summary of responses and other independent contributions
to BIOMCH-L. It contains useful information not only about video camera speed,
but also about various electronical and mechanical features of
video recording systems (if you have not read it yet, I suggest you to read
the wonderful contribution by John Yelland, which was independently sent to
the list, and is included here).

My plea was elicited by the following surprisingly
"naive" statements contained in a recent posting by Dr. Gideon Ariel:

[1]
> Video cameras present new and quite serious
> problems since they exhibit inertial constraints from start up until
> they reach full speed, as well as variability even when their
> maximum rate has been reached.
[2]
> This variability has been demonstrated [NOTICE THE WORD DEMONSTRATED!] [notice here]-------------^
> to be as great as 5 fields per 100 fields, that is a 5 percent
> variation. In other words, the camera might start at 59 f/s go to 62 f/s
> after one second and then go to 64 f/s and back to 58 f/s
[Boys, let's go back to the spring cinecameras]
[3]
> Three video
> cameras operating simultaneously would varied non-linearly with each
> other. This is an inherent problem of hardware which cannot be
> corrected by using Gen Lock on the cameras!!!

Dr. Ariel should be the man who you shoud trust when you buy
the Ariel Performance Analysis System... think about it!

Dr. Ariel, after receiving my plea (together with all of you),
and after exchanging some messages with me, posted to the entire list a
new message (included below) where he included, without my authorization,
two of the messages I sent him.
Both of them contained kind compliments for him. Since he used
them for the clear intent of advertising his product, absolutely contrary
to my intention and will, I feel FORCED to at least clarify one of my
statements (notice the BIG IF!):

> It is still not clear [!!!!] to me how your system works, but
> [...omitted...]
> if [IF!!!!!] you have developed a cheap method to evaluate and correct
> the phase shift between the video signals, and provide a common
> "time base" to put together video signals with other analog inputs,
> such as those from force plates, etc., I believe you did something
> important.

After carefully reading Ariel's previous messages, I was convinced
that his system CANNOT correct for the PHASE SHIFT between the
video signals. In fact, he uses just EXTERNAL LEDS to synchronize the
cameras. However, Ariel seemed "somehow" to state the contrary in his
messages, and I wanted to be diplomatic. My mistake.

This is my original plea to BIOMCH-L:

> Date: Wed, 7 Dec 1994 02:57:44 +0100
> From: "Paolo de Leva - Sport Biom. Rome"
> To: Multiple recipients of list BIOMCH-L
> Subject: videorecording speed variability

> The recent message by Ariel about his system contained some
> interesting notes about large variations of videocamera speed (fields/s).

> Ariel stated that the speed of the videocamera (and not that
> of the tape recorder) is initially lower than the nominal speed (60Hz
> for NTSC standard), then after a given period of time reaches a
> value that, anyway, does not at all stay stable (+- 2Hz!).

> That was completely new to me, and contrary to what I assumed true.
> Unfortunately, Ariel wrote his message mainly with the purpose of
> describing the evolution of his system during the past years, and
> did not fully explain the reasons of the unstability
> (or variability) of camera speed.

> 1) I know that the signal recording starts some seconds after
> the recording button is pressed, and I believed that the delay
> was electronically controlled by a system able to monitor tape speed,
> and that the system did not activate the recording until the tape
> would go up to the required speed..

> 2) I have always assumed that the camera "SAMPLING" frequency
> was controlled by a quartz, threfore as much accurate as the speed of
> a digital watch. Of course, the video signal from the camera is recorded
> on a tape that may run in front of the recording heads at a variable
> velocity, because of mechanical reasons. The above conditions imply that
> the magnetic signals representing the sampled images (fields)
> are physically placed on the tape at variable distances from each other.
> However, kinematic analyses are performed digitizing one field at a time,
> and the fact that reciprocal distances between fields are variable
> should by no means affect the accuracy of the kinematic analysis.
> For obtaining accurate results (if we assume constant camera speed)
> we only need that the electronic shutter of the camera is opened
> at a constant rate (and stays open for a very short interval, to freeze
> the image of fast moving objects..., but that's not the key point here).

> Where am I wrong? The info given by Ariel might imply that somehow
> the camera electronic shutter is controlled by the video
> tape recorder, and its rate of image sampling varies with the physical
> speed of the tape. If this was true, what should be the reason?
> (A second unlikely alternative would be that the camera speed
> is indeed given by the camera itself, and the electronics of the camera
> cannot maintain a constant speed; however, this seems to me a not
> realistic hypothesis)

> Is there anyone among the subscribers (either Ariel himself, if
> this meets his commercial policy, or others) who can help me to better
> understand what are the reasons of the videorecording speed variability
> reported by Ariel?

> With regards,

__________ _________ ___________~___ ________ _________________~___
/ ~ ~ ~ ~ \
/______________~______~__________ _______~_____~______________~_____~_____\
| Paolo de Leva ~ \ Tel.+ FAX: (39-6) 575.40.81 |
| Istituto Superiore di Educazione Fisica > other FAX: (39-6) 361.30.65 |
| Biomechanics Lab / |
| Via di Villa Pepoli, 4 < INTERNET e-mail address: |
| 00153 ROME - ITALY \ deLEVA@RISCcics.Ing.UniRoma1.IT |
|_____________________~________~__________________ __________________ _____|
Panta rei :-)




~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~
Date: Tue, 6 Dec 1994 22:10:54 -0800
From: ariel1@ix.netcom.com (Gideon Ariel)
Subject: Re: videorecording speed variability

Dear Dr. De leva:
The problem is not with the electronics. The quartz is sampling in
a percise 60 Hz. However, the video tape itself is moving mechanically.
This mechanical pull is stretching, some how, the tape. In our
laboratory we video a crystal clock at 60 hz percisely. However, we
found variation up to +- 3 fields per 100 fields. You can try it
yourself with any video camera. The one we used was Panasonics 455
S-VHS. The variation occured at random. Not necessry in the beginning
or the end. In fact the video camera speeded up and slowed down few
times.
To give you another example, try to put pressure on the tape movement
in the video camera. The crystal frequency does not change, of course,
however the video tape slow down. Another word you have different
number of fields per given length of tape.
Please let me know what you find out.

Gideon Ariel, Ph.D.



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Wed, 7 Dec 1994 9:01:35 -0600 (CST)
From: YUB@rcf.mayo.edu
Subject: RE: videorecording speed variability

When I was working in the Sports Biomechanics Laboratory at the University
of Iowa as an research assistant to Dr. James G. Hay, we use two Panasonic
AG450 S-VHS video cameras. A lab technician calibrated both of these two
cameras. The calibration results showed that both cameras had a constant
frame rate of 59.999999.... Hz. I don't know anything about video cameras,
but the calibration results doesn't seem to support Ariel's statement.

Bing Yu, Ph.D.
Orthopedic Biomechanics Laboratory
Mayo Clinic
Rochester, MN 55905, U.S.A.



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~
Date: Wed, 7 Dec 1994 17:25:28 +0200
From: "Technical Services Unit, Carlton"
To: Multiple recipients of list BIOMCH-L
Subject: Video Cameras speed

I was interested to read Gideon Ariel's posting re APAS. The concept of
digitising the video signal direct to Hard-disc and then analysing it will
surely be the way of the future, as the technology to record a reasonable
video rate at acceptable quality matures. Video recorders, for their
designed purpose, are useful and reliable machines, with known limitations.
Unfortunately, as Dr. Ariel pointed out, those limitations become quite a
problem when using standard, broadcast video units in precision
measurement.

However, the comments re frame rates are quite extraordinary.

Speaking as a video engineer of 20 years experience, I assure readers of
this list that video CAMERAS do not vary in speed. They use a quartz
crystal oscillator to set the timing of the horizontal and vertical sync
and, after a warmup pariod of a few minutes, will run at a stable speed for
a very long time. Such is the stability of even basic domestic camcorders
that I have had two camcorders produce tapes that ran in sync. for nearly
30 minutes.
By using one camera or a master oscillator and "gen-locking" the remaining
cameras or video processors, a whole system runs "in-step". The stability
of the master oscillators used in broadcasting is at least 1 in 10^^8.

What does vary is the speed of the video tape.

There are several control systems in a typical recorder. During recording,
the speed of the tape and the speed and position of the video head drum are
servo locked to the incoming video signal. The video signal track is laid
down on tape as a series of evenly spaced stripes, with a sync signal
(control track) derived from input vertical syncs.

When the recorder starts up, the tape goes from rest to normal speed within
one second, and the speed and position servos set the head drum correctly
within 2 seconds, or less. Depending on the type of recorder, and on
whether the tape is blank or not, this period of settling will occur during
blank or pre-recorded tape and recording doesn't start until settling
occurs. After this, the recorder runs at constant speed. It is quite easy
to record 10 seconds of anything before beginning your trial and in fact
this is a standard procedure in video production, to give a stable section
of tape for the "Pre-roll" of the editing machines.

During playback, the tape speed is set by comparing the frequency of
control track pulses with an accurate crystal oscillator (or external
video). Any variation of recording speed is then tracked by this servo
system, giving a constant frame rate output. The control track pulse also
provides a position signal for the head drum ( the "Tracking" control
adjusts this) to put the video heads at the correct place on the tape to
read the video signal.

The effect of mechanical tape speed variation (wow and flutter) is to
slightly vary the replay frame rate (by less than 0.2% in a new machine)
and to introduce time errors in the horizontal TV lines. These "time base
errors" are less than about 1% of the length of a line and are only
important when trying to mix two video tape signals together.

To see the effect of speed errors on your VCR, use an underscan monitor and
look at the bottom of the picture. There is a point about 6 lines from the
bottom where head switching takes place. The image below this point, that
is the signal from the start of the sweep of the head onto the tape, may
have a horizontal phase shift, ie a time error, compared with the signal
above the point which is from the end of the sweep of the previous field.
This time shift is due to errors of tape movement and is usually a small
fraction of a line (eg a few nS).

If the VCR is correctly adjusted, there will be a small "Y" shaped blurring
on vertical images and there may be some small to and fro movement. If the
image appears to be greatly displaced horizontally, the tape tension of
either the recorder or the player is incorrect. Some VCR's eg Umatic have
a "SKEW" control which adjusts tape tension to correct this shift. The "Y"
shaped blurring is also caused by the minute error in positioning the heads
during manufacture. ie they are not precisely 180 degrees apart on the
drum.


John Yelland



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Wed, 07 Dec 1994 11:02:44 -0700 (MST)
From: hinrichs@asu.edu
Subject: Re: videorecording speed variability

Paolo:

I have tested the feld rate of our video cameras/recorders here at ASU by
filming a clock accurate to 1/1000 second. The field rate is steady from the
first video field I looked at. Phil Cheetham (one of the co-inventors of the
Peak system, and also a graduate student of mine) pointed out to me that the
precise NTSC standard frame rate for color video is 29.97 Hz which
translates to 59.94 fields per second. As I understand it, black and white
was originally 30 frames per second exactly. When color was added, they
came up with a number close to 30, but differing for some technical reason
related to the frequency of adding colors. If you hear a more precise
explanation, please relay it back to me. I want to be sure I am teaching
my students properly. I build into my programs the assumption that my data
are being collected at a sampling rate of 59.94 Hz.

Best regards,

--Rick

----------------------------------------------------------------------------
| Richard N. Hinrichs, Ph.D. | email: Hinrichs@ASU.EDU |
| Associate Professor | or Hinrichs@ESPE1.LA.ASU.EDU |
| Dept. of Exercise Science & PE | or atrnh@ACVAX.INRE.ASU.EDU |
| Arizona State University | Phone: (602) 965-1624 |
| Tempe, AZ 85287-0404 USA | FAX: (602) 965-8108 |
----------------------------------------------------------------------------



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~
{This is a message I sent to Ariel after his first response. THIS MESSAGE,
OF COURSE, WAS NOT INCLUDED IN ARIEL's POSTING TO THE LIST: it was not
good advertising, I guess!]

Date: Thu, 8 Dec 1994 00:01:18 +0100
From: deleva@risccics.ing.uniroma1.it
To: ariel1@ix.netcom.com
Subject: Re: videorecording speed variability

Thanks for your answer, but you just repeated what I wrote in
my message: the distance between fields may vary in different sections
of the tape. [...]

I repeat that the accuracy of time measurements in kinematic analyses
depends ONLY on the rate of SAMPLING, and not on the distance between fields
on the tape!
We only need that the images are SAMPLED at a constant rate. The rest
does not matter at all, as far as I can see. That's where I need further
explanation: since you confirmed that the sampling rate is perfectly
constant, what do you need to adjust for, both with the special card
developed for you by the electronic engineer, and with your software?

That was not at all clear before, and it's not clear now as well.
I appreciated your answer anyway, and I would appreciate further
clarification.

With regards and esteem,

Paolo de Leva



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Wed, 7 Dec 1994 16:18:17 -0800
From: ariel1@ix.netcom.com (Gideon Ariel)

[...]
First, let me clear the point that I am not a video expert. I do not
have to tell you that since you already figured it out. However, I will
tell you now where the problems that I encountered:
1. When we display video field from a pre-recorded video tape, each
field is off by one line due to the interlace format. Here, we can be
one field off if you digitize each field and there is a shift of one
line.
2. When we take video session of , lets say golf swing, we have 3 video
cameras at different angles. Each camera can be off within the field
since they are not "gen-locked". If they are "gen-locked" this problem
will not occurred. So here is another discordancy up to one field.
3. When we display the video signal back on the monitor and using the
Panasonic 6300 model, the field advance is either add or even. So, if
we want to extract every field we might loose synchronization between
each of the 3 cameras.
4. When we are using the Panasonic 7350 which can be advance field at
the time, we found that the advance is not consistent. Some time it
will advance on the even field and sometime on the add field. So, the
counter is not reliable. This is another field potential discrepency.
5. When we want to synchronize the video with analog signal, we
utilizing LED light that send a signal to our A/D board and the same
time light the LED lights in front of the cameras. We found there
another one field discrepency.
6. If we use more cameras together in the same session, lets say 6
cameras, we have more chances for discordancy.
I fully agreed with you, since you are the expert, that the camera
signals are accurate. However the procedure to synchronize multiple
cameras with external signal and processing the video tapes one at the
time, in our case, caused problems. The way we solved this problem is
to capture the video digitally to the hard disk and then utilize the
software to synchronize the signals. However, I agree that with
non-interlace cameras, gen-locking mechanism and accurate VCR some of
these problems would be resolved. However, this is an expensive
solution.
How would you resolve the problem of interlace fields discordancy
from field to field.? There is an "up and down" movement when you
display the fields one at the time.
Thank you very much for your comments
Respectfully
Gideon Ariel, Ph.D.


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Thu, 8 Dec 1994 17:15:35 -0800
From: ariel1@ix.netcom.com (Gideon Ariel)
Subject: Re: videorecording speed variability

Dear Dr. De Leva:
Thank you for your message. You are correct, our board can
distinguish between add and even field. That is done, according to my
engineers, by which pixel is used first. Then we can adjust with
software and correct for the deviation.
We also using now 120 hz camera. We shrinking the pixels to half by
a special compression board, then we grabbing the image which made of
two fields vertically. (Look elongated and compress 2 pixels to one).
Then we use software to decompress the image to give a full screen
image. This way we do not loose resolution.
I will be happy to send you images in BMD or PCX file format on ftp
is you are interested.
We conducted here additional experiments to find the frequency of
regular video cameras. You were correct, after the camera started we
could get as much as 1/4 field shift on the oscilloscope. The
discrepancy occurred mostly in the mutching of more then one camera
together without the gen lack. I could not repeat more then two fields
shift. One of them was due to the shift between the interlaced fields.
Thank you very much
Gideon Ariel



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Thu, 8 Dec 1994 17:56:28 -0800
From: Gideon Ariel
Subject: Video Analysis
To: Multiple recipients of list BIOMCH-L

Dear Biomch-L Subscriber:
After corresponding with few of the Subscribers I can clear my
statement about the video fields/second problem.
1. The crystal frequency of the video cameras is extremely stable.
2. In order to digitize manually or automatically 60 Hz video signal
there is a shift due to the Interlace signal . This can result in an
error of 1 field.
3. If data is collected with 2 video cameras, as long as gen lack
is not used, another discrepancy of one field can be occurred.
4. If you rely on the VCR counter, it seems that the counter is not
consistent with the Panasonic 7350 and lack on the even or add fields at
random. This can result in additional one field discrepancy.
5. If the system grab in RAM multi frames, and then after the
digitizing this frames, lets say 16 frames, needs to find the position
on the VCR again, there is another discrepancy of one to two fields.
6. If the user intended to collect analog data such as Fore Plate,
EMG etc. with the video, synchronization may be another source of error
up to 2 fields.
All the above potential problems can be resolved by grabbing the
fields digitally to the hard disk and then process digitally the frames
to correct for potential errors in synchronization. Then synchronization
of external signals such as force plate, EMG, and other transducers, has
to be carefully evaluated and tested.

Enclosed is two of the correspondences I received from Dr. De Leva:

[...omitted...]

I think this issue is very important to discuss since the raw
digitizing data set the tone for the accuracy of any biomechanical
analysis of movement.

Respectfully,
Gideon Ariel, Ph.D.




~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Fri, 9 Dec 1994 10:20:32 -0500
From: Paul Guy
To: Multiple recipients of list BIOMCH-L
Subject: Re: Video Analysis

Dear Dr. Ariel,
In your letter below, I think you are being unduly pessimistic about
the time accuracy that can be attained. We are satisfied with no worse
than about 1/4 to 1/2 frame!

>
> Dear Biomch-L Subscriber:
> After corresponding with few of the Subscribers I can clear my
> statement about the video fields/second problem.
> 1. The crystal frequency of the video cameras is extremely stable.
> 2. In order to digitize manually or automatically 60 Hz video signal
> there is a shift due to the Interlace signal . This can result in an
> error of 1 field.

.....if you use a VCR that will examine fields (such as a time lapse
VCR, ie Panasonic 6750), the frame that the computer sees is really a
composite of the same field, so that what you digitize as a frame is
really a field. We used to use a Sony Motion Analyzer (long since
obsolete) that did the same thing. If you use a shuttered camera at
1/1000 sec., and GEN-LOCK them all together, and derive your analog sync
signals from the vertical TV synch pulses, you can achieve very close
synchronization. We can use video or Northern Digital Optotrak 3D camera
systems together ALL gen-locked, and achieve 1/4 frame or better
synchronization between video, 3D coordinates, and analog data.
Although such tight synchronization may seem a bit crazy, all the
collection systems can effectively 'shutter' their processing in full
synch to get this performance. We had some temporary problems at one
point when we were doing an assessment of efficiency of a prosthesis
that replaces the lower leg and foot. At 1/4 frame increments from true
synchronization, the efficiencies went from 70%,120%,150% and going the
other direction of synch, they similiarily decreased. All this was done
at a sampling rate of 60 Hz. Needless to say we were highly motivated to
fix the synch up!

> 3. If data is collected with 2 video cameras, as long as gen lack
> is not used, another discrepancy of one field can be occurred.
> 4. If you rely on the VCR counter, it seems that the counter is not
> consistent with the Panasonic 7350 and lack on the even or add fields at
> random. This can result in additional one field discrepancy.
> 5. If the system grab in RAM multi frames, and then after the
> digitizing this frames, lets say 16 frames, needs to find the position
> on the VCR again, there is another discrepancy of one to two fields.

....hmmmm what system are using for this? It sounds like the PEAK
measuring system.

> 6. If the user intended to collect analog data such as Fore Plate,
> EMG etc. with the video, synchronization may be another source of error
> up to 2 fields.

...absolutely not.... you have severe problems with the way you are
collecting analog data if this is the case. If you are really stuck,
sample your analog data at least 4 times the video frame rate, and use
the vertical synch signal as one of the analog inputs. With the added
start/stop from the rest of the system, you can achieve very close
synch.

> All the above potential problems can be resolved by grabbing the
> fields digitally to the hard disk and then process digitally the frames
> to correct for potential errors in synchronization. Then synchronization
> of external signals such as force plate, EMG, and other transducers, has
> to be carefully evaluated and tested.
>
....true, but what a painfull and expensive way to go! You CAN do it
on the cheap, but you would need some minor electronic addons.

> Enclosed is two of the correspondences I received from Dr. De Leva:
...[deleted, they were mostly queries].....
>

I can offer some help in these areas, as we have been fighting the
synch problem for some 15-20 years, with a very wide range of new and
ancient equipment. It's not that difficult, but when you don't have a
good grasp of the electronics and video stuff, it can be a really alien
mind game.

-Paul