Need help with understanding video encoding.

[andilovellt]

Hi. I would like to ask video expert's opinions.

My current work has to do with videos. My task is to process footages of chemical reactions that has been submitted to me.
What i do is sync the time of the experiment start or the adding of the reactants to a player that allows me to navigate the video frame by frame. My task is to time the distinc chemical occurances that is happening at a progressive chemical reaction. I should provide time accurate to 1/100 of a second.

I had been working on PAL videos about one to three minuites long and at my player, each frame i move is like 0.04 seconds. With this, the margin of error is supposed to be +-0.03
I had been experiencing some of the videos have duplicate frame (when i play 1 frame forward, there are times that i get an exactly identical frame as the previous one)

My questions are:

1. How should i treat duplicate frames? Are they inserted to make for the time without video coverage or are they like a random
thing that pops up causing additional time in between twon occurances within the video?
2. Should i be counting in or should i subtract the time elapsed by the duplicate frame?
3. If i have had 3 duplicate frames in between two occurances, does that mean that the next time is actually off by 3*0.04 sec?
4. What causes this duplicate frame to happen?
5. If frame duplicates where actually bad, is there a way remove the frame duplicates without causing an alterations to the actual time?
6. What would i be expecting if my Video is NTSC?
7. Is there any video format that natively has 100frames per sec framerate?


Here is the properties of the video im currently working at:

Format : AVI
Format/Info : Audio Video Interleave
Format settings : rec
File size : 15.2 MiB
Duration : 1mn 30s
Overall bit rate : 1 406 Kbps
Video
ID : 0
Format : MPEG-4 Visual
Format settings, BVOP : Yes
Format settings, QPel : No
Format settings, GMC : No warppoints
Format settings, Matrix : Default (H.263)
Codec ID : DIVX
Codec ID/Info : Project Mayo
Codec ID/Hint : DivX 4
Duration : 1mn 30s
Bit rate : 1 305 Kbps
Width : 352 pixels
Height : 288 pixels
Display aspect ratio : 1.222
Frame rate : 25.000 fps
Standard : PAL
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Bits/(Pixel*Frame) : 0.515
Stream size : 14.1 MiB (93%)
Audio
ID : 1
Format : MPEG Audio
Format version : Version 1
Format profile : Layer 3
Codec ID : 55
Codec ID/Hint : MP3
Duration : 1mn 30s
Bit rate mode : Constant
Bit rate : 96.0 Kbps
Channel(s) : 2 channels
Sampling rate : 44.1 KHz
Video delay : 24ms
Stream size : 1.04 MiB (7%)
Alignment : Split accross interleaves
Interleave, duration : 2923 ms (73.06 video frames)


Thanks!

[pandy]

not sure what type environment You use - i assume that improper tools make some limitations - what is format for footage send to You is that JPG or similar pic from fast video camera (ordinary video cameras don't go over 60fps - for faster than this You need or special camera or camera that is able to capture video faster than NTSC)

Try to use AviSynth, most of the tasks described by should be possible to perform with help of the AviSynth

[Nelson37]

There are most likely not any duplicate frames, this is probably a minor error in playback.

Need more info on the camera being used. Also software.

See if you can change the compression used. Uncompressed AVI or MPG I-frame would most likely eliminate your problem.

MPG-4 is, IMO, the wrong tool for the job.

[pandy]

compression is not a problem - MPEG-4 can use only I frames like MJPEG or MPEG-2 - this is not an issue. most important is framerate of the camera that is the source for footage - if camera is not able to record with 100 frames per second then whole idea difficult to perform - PAL cameras usually record 25 frames per second - interlace cameras 50 fields per second, NTSC are 29.97 frames or 59.94 fields - half of required time resolution.

[jagabo]

Frame rate : 25.000 fps
Scan type : Progressive

pandy is correct. You cannot get a 1/100 second precision out of a 25 fps video. And without knowing the capture device and software used it's hard to say what duplicate frames mean.

[pandy]

Only one thing is not clear for me:

I should provide time accurate to 1/100 of a second.

maybe 25.0 fps is enough? - time can be provided with 1/100 second accuracy ie each frame occurs 40ms thus we are able to represent time of reaction with 4*1/100 sec (assume no jitter between frames so 0, 40, 80, 120 ms....)

i'm really confused now...

[jagabo]

Only one thing is not clear for me:

I should provide time accurate to 1/100 of a second.

maybe 25.0 fps is enough? - time can be provided with 1/100 second accuracy ie each frame occurs 40ms thus we are able to represent time of reaction with 4*1/100 sec (assume no jitter between frames so 0, 40, 80, 120 ms....)

i'm really confused now...

You can also say that 1/25 second is 40,000 microseconds. That doesn't mean you have microsecond accuracy.

[pandy]

but You can provide time even with ns accuracy... so it is not clear to me - jitter below 1/100 sec or 1/100 resolution - jitter bellow 10ms should be easy to fulfill by camera (i expect jitter bellow single us even for simple analog cameras) but acquisition can introduce jitter.

time resolution 10ms require at least 100fps (or 50 fps interlace thus 100 fields if vertical resolution is not important) camera.

[jagabo]

The problem is, with samples every 1/25 of a second you can't tell the difference between 1/25 second, 1/26 second, 1.1/25 second etc. It's a matter of "significant digits". You can label your axis 40 ms, 80 ms, 120 ms, etc. But that doesn't mean your measurements had ms precision. If the OP needs 1/100 second accuracy he needs a camera that shoots 100 fps. In scientific reporting, if you say something took 40 ms you mean it took 40 ms -- not 39 ms or 41 ms. If you say it took 40,000 us sec you mean it didn't take 39,999 us or 40,001 us. Ie, the number of digits you report implies the precision of your measurement.

[pandy]

I know that - my point is that i don't know what are expectations - accordingly to topic creator - He need to provide time with 1/100 sec accuracy and 40ms is accurate to 1/100sec - so im confused - it is obvious that without 100fps camera no one is able to provide 1/100 sec time resolution so maybe issue is somewhere else - this is not clear.

[jagabo]

I guess we'll have to wait for the OP to indicate exactly what his needs are.

[Gavino]

maybe 25.0 fps is enough? - time can be provided with 1/100 second accuracy ie each frame occurs 40ms thus we are able to represent time of reaction with 4*1/100 sec (assume no jitter between frames so 0, 40, 80, 120 ms....)

i'm really confused now...

You certainly are.

Let's say you see something happen in the second frame.
All you can say is that it occurred sometime between 40 and 80 ms from the start.
You obviously do not have 1/100 sec accuracy.

[pandy]

ok seems that we need to remember that resolution is something different than accuracy - i can sample time 25 times per second with accuracy 1/100 sec ie 40ms +-5ms this not mean 10ms resolution.

So im confused what thread author mean when he says

I should provide time accurate to 1/100 of a second.

[Cornucopia]

Look at these 3 different sampling styles:

 ___|___|___|___|___|
+---+---+---+---+---+

 __||__||__||__||__||
+---+---+---+---+---+

 _|||_|||_|||_|||_|||
+---+---+---+---+---+

All three have the same "sampling period" (+ sync marks), but each has a different width sampling window (| marks).

Thus, just like camera "shutter speeds" can be different from frame rates, you have to choose which requirement(s) you need. You could conceiveably set up a pair of cameras (using a beamsplitter) that are genlocked, but offset by 1/2 the framerate, so that you end up with a framerate double what is normally capable. However, their time resolution might still be crap, depending upon whether you can set their shutter speed to be shorter than their framerate or not.

Scott

[olyteddy]

Stick a stopwatch in the frame and visually interpolate?

[pandy]

yes, using external reference source and sync to this source - one of workarounds that can help, also multiple cameras can be used and synchronized during post-processing, also source of flash light (stroboscope) can help to solve some issues. Still we don't know what kind of setup is used for making footage.