Cycling '74 Forums

All MaxMSP Jitter Dev Javascript Java Jitter-Dev Max For Live Misc Edge

How to save frames when they differ from a previously recorded frame?

→7 posts by 4 people

→Started 1 month ago by gpvillamil
→ Latest reply from gpvillamil

Tags:

No tags yet.

gpvillamil

I'm trying to process an incoming video stream, and only record frames when they differ by a specific threshold from the last recorded frame. I've found the jit.change object, but that only compares against the previous frame.

What I want to do is basically:

1) Get a frame
2) Compare it to the last recorded frame
3) If the difference exceeds a threshold, write the frame to the output
4) ... and make it the last recorded frame

I think this should be pretty easy, but I'm kind of running around in circles with it now.

1 month ago

Jesse

Something like this?


----------begin_max5_patcher----------
677.3ocyWssiaBCD8YxWgkUeLEgM26SseGUqVY.ua7JtjBNMoc09uWyXHW5l
z0PHz7hQyXa3Lm4LiMutvBmTsi2fQeA8cjk0qKrr.WsNr5rsvErco4rFXY3x
MEI7Z7R8TxZQSJKm2Nkicbm2mpJkMheCdIDa+XeRTX2bp8KJy4R3sQN3rZir
2Ksyq1k7Wq4Z.hEkR7RDNgU9LF8P2pVyjoqDkO+XMOUpWnquisyRDwOr8g1f
pF2umV.VxJf2K9a0BVde.Ix.eUIu7YRHt02aKVzNrzP9of2zvdlumf36.Pgk
qp4MqPehfGJIQOKIQtDIYDyDQAlIVaLXlIZLLSxForpDatN3hg3GJ.HTHxnA
5TeP+398bbvDLko4ZljeojbKmeiyuzXHwFBwsWzHSudigQdJuREACIvGX0u5
CvLq9uiE7HWU4OcR0EbV1+MQAA5CFAzgq+HEENigNJ4aUa9crwKBo8Oj1EU+
TvugJF0m4wBl5Lpcsxl+cGi.fiHDfdBhFml4tuunNJccgwv9wy0VbTG90lWW
uUTlUsEOA4OSyZt5NcQPkdfy4CH+IT+JQ4njanv0ntb8R1.c1jZ6OFIq2DWV
qq2PeUcemMHm6ihaZDvTzPfoHNjwUc6NwTU5JUJ99p+G0EN4LfNxyLOlgfsh
yEk+8OX.e9V+mRaMUapS6gd20LWhNffLdiTTxjBUyxiVzIqYkHKiCS2yeEhr
0Upebno+BvwPchpOu5AILXuE5gylcMFrgl.1ngg1qESl.of4ERdFfoYlknlP
SdyJlBu6XISpEmWNRciz6NojAHxeVQj6z2BkFCMMcc7fCI7faLepEkBV8sWa
stx3nUZQ9f3vcXwAIt6OCi.PpiJvZBv5Dy4yftbnrWWtMPeYp3CVmfUkwaK9
CzmXoRH
-----------end_max5_patcher-----------

1 month ago

gpvillamil

That's helpful, but doesn't seem to work with my footage.

Basically, I have footage shot from a moving train. Whenever the train comes to a stop, I want to skip frames, until it starts moving again.

The footage is here: http://www.villamil.org/video/train.avi

What I'm finding is that lots of frames are being passed even during the stops. I suspect that this is because of camera noise, lots of cells appear to change from frame to frame?

Ideally I want this to work as fast as possible (not real time) - sending a framedump message the jit.qt.movie only seemed to go at the original movie's frame rate.

I'm thinking that subtracting the new frame from the last recorded frame using jit.op @absdiff and then taking a jit.3m of the result might give a more accurate indication of difference, and the use that to trigger a gate?

1 month ago
laserpilot

Yeah I would say you definitely need to do some sort of tolerance gate with jit.3m or jit.op involved

When I made a light painting patch I used jit.op @op - with a very small value to help lower the camera noise in certain color channels (using jit.unpack) in order to lower the amount of noise going into the feedback/painting patch...if you get it working it would be a useful patch...I know I've heard of similar things with cutting up video files based on cuts and it would output individual files.

Also, you could try using jit.slide or jit.glop on your abs diff output..they would act as a sort of low pass filter and filter out the higher freqency noise involved in the camera sensors with the right settings...

4 weeks ago

gpvillamil

Here's what I came up with. The little change detector subpatch calculates a change value between successive frames, and if it exceeds a threshold, then it writes the current capture frame.

The advantages over the jit.change method are that 1) it is measuring the amount of change, not just the # of cells that have changed, and 2) it makes available the change value, which helps when setting the threshold.

I've put a little graph of the change values in the patch. You can see how if you move really slowly, the difference gradually increases until it trips.


----------begin_max5_patcher----------
1972.3ocyassiiZDD8YuR4e.gxiNinuQC4gnMJOkugUQivPaO8tXvAZOyrIJ
+6ou.d.iwtwKfYkF.SeqpSUcUEUW6+9SeZk6l72YktN+pyWbVs5ekuYk9cp2
rp9Eqb2G8dbZTotit6OlJ3ko7DVg65pNTxDkhumxTc.U+xri6yOJRkModMr9
0wQYIb4EluYc8dhhT+as7IODAFhUOQCBfHxZGvSdN+U8PODIhegms64BVrvL
ZZfbTqc7AD0MDHPcC340bXFpP78CLyXbcWK+6i1OQQjZJhDfQ.McD.oXBP8j
7MHJ5LJRx364YR7odjJh.1rGmlbbcWvnPHvWMk9fPeu.MiS8HAAmytrnuEmm
lWbZjgAdnv0W7IvEWTT8PggdxESO..f5SzXrjgf9c4nZjxqyzEbhM+.VvRRW
SID0J.Na53IZkl7Me8WP9tclPX8DBvDuPMT.krCDpdBQ7C7OGwkZU7rZkJPm
IjVOgATkLTOgZI25FKhdB0i7+9oOotKus15M.h7c6jp5WVMGbEcUHQqjh0Wg
d0WuhhJOS3dYrDeZ8OTvJYYhHAOOqwh46S0Kfdw7ptzGLd2XQF6MI4bhVDr2
EFiDLQQtCBcpks4YhR9+nYLfbCx8fdHEWP0LDIvBzaST1tFvmhDxh1qa082K
3QotWBYI2FYMjfMHK7tQ18rxxncrtPaoH+vHBqxsDJVHzX5RCwJinWy54fgT
j67ANQEhwDc7HMPG7jfNfYCcJXQIiH3f7ap5PlBvAFN1fSeVr9JW7zeKdZe9
qb1.wH30vHjFc.Fy891X1RRIOuORTve+r3TrDxntirQ9so4xYZDAEeZkiPMn
PrwSnjFhDJ73tLpCus6RRnVKNH3g3uju04m2Bb9M4Uni3EVlilOGu8p9dFrF
ToNhs.yGNLO5F5GeUOCNLipdP2YRGRYAyX2vQ1WgymEuTbzwa7AuS1zPSu0L
P3bAeBmTmMiokeiQNPHo5iBI2dC28pjAnyEJcvI9EIExbRXBIqlWL91nPl.J
.dzovFEl3d8OkTSXrhJvoFcjSOOk8JqnT52nY+W41HZQrdGAhp8k3iMlZBZx
ARfeWZd72XIM+B6UtIrsCbdxOvx9XHdMca012koq7rld95r3QGSEO2iHrcG1
FEy5e3WV.rxcWAOIOSQHsGq580K4WbLYwQcsA0q6RVzgKMbQdd5lnhW4k7Ml
zO8gbQpgGkwk1dXBtgnfdeLR99CE5Out4zwxhjSyKkwE4oosmMSSudolRXux
iYuwSDunmtF.Wq7q0Tcp0Nv1Mb0chcM4i12ps9jgc2Jha1zE1NV4bjB0dJC6
X+5J4Va844X6FaP6XOy28L59baZq9vtV0iqmDDV4UHse.N3Z.L7V.LPqsiCL
QdZPa6.36EXQKEbUo4lev4yx+h1Tlv2tcZA4Pe8MerknbegsLTEYP+.Nb1Ab
iq6oEnM9vCmabFtjTrK1sAldbezjhzPn4rN.nYFpoyITqma29wJfUXkIBJSt
vgjaahsc6Mid7JFPaIQiy2umYBuv0czQECEeEXw6FvhuI.SB7Z3RS9FeEg9P
46Fw1kxy5M9HMYq5POHRY9wh3ZQVkyMmyXgDVofmcJj2u7gUYGP6d9BOI4rn
K0vNO4PtLNwJhr5XirRVNX5WSV1w.zt8bAv.ntv5UD.KO5W4EwNA.ZYR+TqU
f7WjL.19svvkoDv21s.JKpiLCT2fhWpOM+nWYIOKWOok+miDRm+aNJLVbamb
ia9g8cas2rErKMeSTZUdTNETg6fxWvMSIwEF7It+ty5Us2qNo85OhNHNVvRb
JEErndRHcPeI8x61EvBxj5CHn6mA5tUFOYU9QzQ707xfyTHtEy0n7Rt1b2yg
lPnlSDkRmzCMoWoxelc3nX7EI.exbJRBuGIxHAsyUUFTkZWp4ThmlxL.DLWm
UrNygNw4Ir3xABRWSuy2TYPAlJKgR5l9aKABe24RMYGSnggTd4XV1EUkyG0r
QjNI0cAfLykVfj8xKRFwSXhXTTBLe+F0aFNMtQ+vLUXyg23YI4u4NbDn14It
5L1LP.dPU+4EyV2XwdxnsDxv9F9NfJYKD3YcwBd1IG1ju7d.mAcrIjoI3Xnq
wFyM.lL8J9AiM.d2EQJ9T599QqgzY6LimfhAEqiHhPm1ZAcz8P7iYsqNtzQx
ZGdNMJj79S6JhFy5qnpj5.U+G.XNpEkYKrJ8AcOd6V.sphU5jTnlyVQrFmlW
xFSvA2DbBmjR7c1.m2J3BlCx6ImCujKx+5A1Nmr7h8MnqebDiDTEvYU0Z.mj
vysDzLya6CRnBDOO6c0PYmr1AamLw9x4a6bldgr00Sl5tnD1dhiZCwcVpeWV
DG9gPaHjMzF7w.bHvBl3TUc+xE4HKYhCakkDxiQrZ0FB0GKNLhy2SWbXDjIn
bScIz9WfPSUj4iU8T+qQgkzl8.2jkBGJKgBMEkkmNrxJFT+qwirusYyAqlfB
z3KpRDDT8gBs9kIuh0rDkLRrjc9POWEbtz7wVSbfYm3v1YPC9PPNP3znoNJD
WfU1yBeL.mUawCdLz1RNhRqnsGiy8kKoAVtggSVtnlUaQeLl1NOJrQx7QU.X
PSAQ17WynAYvB1YwCxKKZISbjkLwAs5C.dLotALW5bpWHu8+.kRzX+.
-----------end_max5_patcher-----------

4 weeks ago

MIB

interesting idea. gonna keep that in the back of my mind :)

just two questions/observations: why do you convert to luma in your "change detector"? Isn't that an unnecessary step? Shouldn't the mean be the same regardless?
Also, and I know this is very minor, the "if" statement is very, very slow. You should probably use a [>] in conjunction with a [sel 1]. It doesn't save you much but according to some quick and dirty bench-testing, [>]+[sel] is more than twice as fast on my machine. Again, not gonna make a big difference probably, but something to be aware of.


----------begin_max5_patcher----------
549.3oc2V1saaBCEG+Z3ovxZWxp3XCLxtnR84XpZhDbZ7DXhBFsrV028gO.o
zIZvrPIUkKrk+.e9e94iO1O45PWWbTTRIem7ChiyStNNXWlNbZa6PySNtIKo
DmFcckVWnndMCopxKpzYBMNHz169D8lcR0C+7fXitYwYq3236QB.SIyuqjbe
6uzrJ5+rWzLe55D0CzSCKSQiWr9WecUOaKUmLsoumccMEdV5JayJpWitkaag
RqRxQ6Su6fLIaXmjMpSBAwlpPzKg3y4k0RHQS8Ni6B98EXo7Q7WAyZOSXPI9
csg5rhVbD8FpVlKN7N.GVfohym.bFFLvT.Ca9.S0ixZyfeSEO7yfmXjKeiYJ
iBGmNXDi2K0RkdXNwdu4z+YJA.PujGdwoDh+3lRn0Ig.17kRH5SRFgN1.wyS
FgfOUID.+HLUPzbmPfekvToHi.yX3Cm2ly.IDmcSnsDZPrvtRX4VBDMUp.iR
EHFrkJu4EGWIfH2R9xVfX3BQuSnHMaayFgfPVOBA1byyvALCePhMddXb0nYR
0+9zaTsl9eM0JKpNroSKc2oSXmDUpnTKUIZY8sv8lT8bdwm1ISSE3vcxKWlt
undyuUCj6GbKzZI8Zy8FRxeQ0zJajzBiIyNGXil.60DGhvi49MOstWqk.frE
EfbKh7MmLWPIEaAk3KNjFKHKXZwX.O1DUsBeqVuFW54AKXG6C2tYvhpn.KTT
zEnn5FO69W.Sm8AZ
-----------end_max5_patcher-----------

4 weeks ago

gpvillamil

Thanks for the tip re (> ) vs. (if ), I've gone ahead and used that.

I convert to luma so that I only have one number to work with. If I take a jit.3m from an RGBA video, it spits out a list of numbers, one for each plane, which I have to process.

3 weeks ago

RSS feed for this topic

Reply

You must log in to post.