Send Motion-user mailing list submissions to
motion-user@lists.sourceforge.net
To subscribe or unsubscribe via the World Wide Web, visit
https://lists.sourceforge.net/lists/listinfo/motion-user
or, via email, send a message with subject or body 'help' to
motion-user-requ...@lists.sourceforge.net
You can reach the person managing the list at
motion-user-ow...@lists.sourceforge.net
When replying, please edit your Subject line so it is more specific
than "Re: Contents of Motion-user digest..."
Today's Topics:
1. Motion in low light (Rusty Lemur)
2. Re: Motion in low light (tosiara)
3. Slow privacy mask loops? (Pascal F. Martin)
4. Re: Slow privacy mask loops? (MrDave)
----------------------------------------------------------------------
Message: 1
Date: Tue, 28 Feb 2017 22:33:59 -0600
From: Rusty Lemur <rusty.le...@gmail.com>
Subject: [Motion-user] Motion in low light
To: Motion discussion list <motion-user@lists.sourceforge.net>
Message-ID:
<CAFzrdHuqMK+M6C+n4tR=0whuktkur-b5oc9gwsjpg-am48p...@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"
Hello experts,
Is it possible to improve performance in low-light scenarios over the
default configurations? I have motion running with a simple USB webcam, but
when the room it's in gets dim, there is just darkness displayed. I'm not
expecting to be able to see in the dark with it, but are there settings I
can use to squeeze as much low-light detail as possible?
Thanks,
Rusty
-------------- next part --------------
An HTML attachment was scrubbed...
------------------------------
Message: 2
Date: Wed, 1 Mar 2017 21:33:17 +0200
From: tosiara <tosi...@gmail.com>
Subject: Re: [Motion-user] Motion in low light
To: Motion discussion list <motion-user@lists.sourceforge.net>
Message-ID:
<cachtdwrcywz_nrwnu3ns1wz772usxr_yqgvu9vdi-oyb6r3...@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
If your camera allows, remove IR filter from lens - this will improve
a little. Want more? Get IR LED light and you will see in the dark
On Wed, Mar 1, 2017 at 6:33 AM, Rusty Lemur <rusty.le...@gmail.com> wrote:
> Hello experts,
>
> Is it possible to improve performance in low-light scenarios over the
> default configurations? I have motion running with a simple USB webcam, but
> when the room it's in gets dim, there is just darkness displayed. I'm not
> expecting to be able to see in the dark with it, but are there settings I
> can use to squeeze as much low-light detail as possible?
>
> Thanks,
> Rusty
>
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, SlashDot.org! http://sdm.link/slashdot
> _______________________________________________
> Motion-user mailing list
> Motion-user@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/motion-user
> https://motion-project.github.io/
>
> Unsubscribe: https://lists.sourceforge.net/lists/options/motion-user
------------------------------
Message: 3
Date: Sun, 5 Mar 2017 20:37:34 -0800
From: "Pascal F. Martin" <pascal.fb.mar...@gmail.com>
Subject: [Motion-user] Slow privacy mask loops?
To: motion-user@lists.sourceforge.net
Message-ID: <dc552463-ade6-ed0b-6c8a-e3e0c4c53...@gmail.com>
Content-Type: text/plain; charset="utf-8"
While searching for something else, I found that function
mlp_mask_privacy (in file motion.c) executes a slow double loop with a
lot of invariant re-calculation in the inner loop:
for (indxrow = 0; indxrow < cnt->imgs.height; indxrow++) {
for (indxcol = 0; indxcol < cnt->imgs.width; indxcol++) {
cnt->current_image->image[indxcol +
(indxrow*cnt->imgs.width)] |= cnt->imgs.mask_privacy[indxcol +
(indxrow*cnt->imgs.width)];
}
}
The operations are for "unsigned char". The current_image and
mask_privacy have the same layout: a flat array of bytes. A single loop
would do at a lower cost. What tickled me was the code comment: "This is
done strictly for processing efficiency to lower cpu". Oh, well..
The following code is equivalent, but more efficient:
int index = cnt->imgs.height*cnt->imgs.width;
unsigned char *image = cnt->current_image->image;
const unsigned char *mask = cnt->imgs.mask_privacy;
while (index > sizeof(unsigned long)) {
*((unsigned long *)image) |= *((unsigned long *)mask);
image += sizeof(unsigned long);
mask += sizeof(unsigned long);
index -= sizeof(unsigned long);
}
while (--index >= 0) {// process the last few remaining bytes, if
any.
*(image++) |= *(mask++);
}
The relative results (with or without gcc optimizations) are below:
$ cc imagespeed.c -o imagespeed
$ ./imagespeed 10000
Testing privacy mask loop (repeat 10000 times)
Execution time (original loop): 25 sec.
Execution time (improved loop): 3 sec.
$ cc -O2 imagespeed.c -o imagespeed
$ ./imagespeed 100000
Testing privacy mask loop (repeat 100000 times)
Execution time (original loop): 64 sec.
Execution time (improved loop): 4 sec.
Doing a flat loop with minimal calculations about double the speed.
Preceding it with the "unsigned long" loop multiply the speed by an
additional factor of 8 on my 64 bit PC. To my surprise, the gain is much
lower (but still dramatic) when compiled without optimization.
I attached the source file of my performance test. It simulates the same
struct indirections as used by motion.c and starts by generating a
binary pattern for the current image and another for the mask,
calculates the masked image using each method and compare the binary
result. If the check passed, it runs the performance tests. That is the
full extend of my testing thus far.
Comments are welcome. I have prepared a modified version of motion.c,
but I still have to test it.
Thanks.
Pascal.
-------------- next part --------------
An HTML attachment was scrubbed...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: imagespeed.c
Type: text/x-csrc
Size: 2895 bytes
Desc: not available
------------------------------
Message: 4
Date: Sun, 5 Mar 2017 22:05:22 -0700
From: MrDave <motionmrd...@gmail.com>
Subject: Re: [Motion-user] Slow privacy mask loops?
To: motion-user@lists.sourceforge.net
Message-ID: <a39d83ff-6af9-df9f-7be9-0d00d5a00...@gmail.com>
Content-Type: text/plain; charset="windows-1252"
I have set up a new issue on github for this.
On 3/5/2017 9:37 PM, Pascal F. Martin wrote:
>
> While searching for something else, I found that function
> mlp_mask_privacy (in file motion.c) executes a slow double loop with a
> lot of invariant re-calculation in the inner loop:
>
> for (indxrow = 0; indxrow < cnt->imgs.height; indxrow++) {
> for (indxcol = 0; indxcol < cnt->imgs.width; indxcol++) {
> cnt->current_image->image[indxcol +
> (indxrow*cnt->imgs.width)] |= cnt->imgs.mask_privacy[indxcol +
> (indxrow*cnt->imgs.width)];
> }
> }
>
> The operations are for "unsigned char". The current_image and
> mask_privacy have the same layout: a flat array of bytes. A single
> loop would do at a lower cost. What tickled me was the code comment:
> "This is done strictly for processing efficiency to lower cpu". Oh, well..
>
> The following code is equivalent, but more efficient:
>
> int index = cnt->imgs.height*cnt->imgs.width;
> unsigned char *image = cnt->current_image->image;
> const unsigned char *mask = cnt->imgs.mask_privacy;
>
> while (index > sizeof(unsigned long)) {
> *((unsigned long *)image) |= *((unsigned long *)mask);
> image += sizeof(unsigned long);
> mask += sizeof(unsigned long);
> index -= sizeof(unsigned long);
> }
> while (--index >= 0) {// process the last few remaining bytes,
> if any.
> *(image++) |= *(mask++);
> }
>
> The relative results (with or without gcc optimizations) are below:
>
> $ cc imagespeed.c -o imagespeed
> $ ./imagespeed 10000
> Testing privacy mask loop (repeat 10000 times)
> Execution time (original loop): 25 sec.
> Execution time (improved loop): 3 sec.
> $ cc -O2 imagespeed.c -o imagespeed
> $ ./imagespeed 100000
> Testing privacy mask loop (repeat 100000 times)
> Execution time (original loop): 64 sec.
> Execution time (improved loop): 4 sec.
>
> Doing a flat loop with minimal calculations about double the speed.
> Preceding it with the "unsigned long" loop multiply the speed by an
> additional factor of 8 on my 64 bit PC. To my surprise, the gain is
> much lower (but still dramatic) when compiled without optimization.
>
> I attached the source file of my performance test. It simulates the
> same struct indirections as used by motion.c and starts by generating
> a binary pattern for the current image and another for the mask,
> calculates the masked image using each method and compare the binary
> result. If the check passed, it runs the performance tests. That is
> the full extend of my testing thus far.
>
> Comments are welcome. I have prepared a modified version of motion.c,
> but I still have to test it.
>
> Thanks.
>
> Pascal.
>
>
>
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, SlashDot.org! http://sdm.link/slashdot
>
>
> _______________________________________________
> Motion-user mailing list
> Motion-user@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/motion-user
> https://motion-project.github.io/
>
> Unsubscribe: https://lists.sourceforge.net/lists/options/motion-user
-------------- next part --------------
An HTML attachment was scrubbed...
------------------------------
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, SlashDot.org! http://sdm.link/slashdot
------------------------------
_______________________________________________
Motion-user mailing list
Motion-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/motion-user
End of Motion-user Digest, Vol 130, Issue 1
*******************************************
