fswebcam consistent image capture

lazardo · 01-10-2019, 07:49 PM

After spending way too much time on this for a development project, maybe someone else will find it useful.

Cheers,

Code:

$ v4l2-ctl -D
Driver Info (not using libv4l2):
	Driver name   : uvcvideo
	Card type     : UVC Camera (046d:0819)
	Bus info      : usb-0000:00:1d.7-2
	Driver version: 4.1.39
	Capabilities  : 0x84200001
		Video Capture
		Streaming
		Device Capabilities
	Device Caps   : 0x04200001
		Video Capture
		Streaming

Code:

$ v4l2-ctl -L
                     brightness (int)    : min=0 max=255 step=1 default=128 value=86
                       contrast (int)    : min=0 max=255 step=1 default=32 value=43
                     saturation (int)    : min=0 max=255 step=1 default=32 value=0
 white_balance_temperature_auto (bool)   : default=1 value=1
                           gain (int)    : min=0 max=255 step=1 default=0 value=33
           power_line_frequency (menu)   : min=0 max=2 default=2 value=2
				0: Disabled
				1: 50 Hz
				2: 60 Hz
      white_balance_temperature (int)    : min=0 max=10000 step=10 default=4000 value=10000 flags=inactive
                      sharpness (int)    : min=0 max=255 step=1 default=53 value=51
         backlight_compensation (int)    : min=0 max=1 step=1 default=0 value=0
                  exposure_auto (menu)   : min=0 max=3 default=3 value=3
				1: Manual Mode
				3: Aperture Priority Mode
              exposure_absolute (int)    : min=1 max=10000 step=1 default=166 value=418 flags=inactive
         exposure_auto_priority (bool)   : default=0 value=1

Code:

ROTATE=-1.57
TTE=68x64+147+58
TRH=56x50+150+127
ETE=44x46+117+206
ERH=46x45+193+207

FS_BRIGHT=34
FS_CONT=17
THRESH=18

fswebcam --quiet -d v4l2:/dev/video0 --no-banner \
	-F 15 -S 15 -D 1 -p YUYV --png 9 \
	-s brightness=${FS_BRIGHT}% -s contrast=${FS_CONT} - |
convert - -rotate $ROTATE -threshold ${THRESH}% \
	-colorspace gray -background white \
	-alpha remove -alpha off +dither -posterize 2 -flatten - |
tee \
	>( cat > envchk.raw.png ) \
	>( convert - -crop $TTE tte.png ) \
	>( convert - -crop $TRH trh.png ) \
	>( convert - -crop $ETE ete.png ) \
	>( convert - -crop $ERH erh.png ) \
	> /dev/null

for i in {0..10}; do
	[[ -s tte.png && -s trh.png && -s ete.png && -s erh.png ]] && break
	sleep 0.33
done

for i in tte trh ete erh; do
	ssocr -C -c digits -d -1 -T $i.png || echo "bad value"
done

Details:

Code:

fswebcam:
-d device		# in this case, Logitech C210 USB and a laptop built-in
-F framerate		# camera and region specific
-S skip frames		# USB cameras need settle time 
-D delay		# USB cameras need delay also
-p camera palette	# 'auto' didn't work
--png 9			# use png over jpg
-s brightness		# adjust for light conditions (don't use 'auto')
-s contrast

Code:

convert:
-rotate			# used gimp 'rotate' tool to find value
-threshold		# adjust value for image (don't use 'auto')
-colorspace gray	# these manage png files and remove color artifacts
 thru
-flatten

Operational theory:

1. fswebcam parameters grab a consistent image. Output piped to:
2. convert (imagemagick) to process image. Output is piped to:
3. tee for parallel processing:
3a. save primary and secondary images
3b. crop four individual numbers
4. sleep delay required else incomplete images.
5. Use 'ssocr' to translate image segments to usable values
6. If 'ssocr' doesn't return correct values, play with brightness, constrast and threshold first.

BW-userx · 01-12-2019, 08:57 AM

I am getting just an white image, so I looked up the commands for imagemagick and added -channel played with the threshold and now I am getting some black in the images.

Code:

FS_BRIGHT=34
FS_CONT=17
THRESH=12
dname=webcam-shot.png

fswebcam --quiet -d v4l2:/dev/video0 --no-banner \
	-F 15 -S 15 -D 1 -p YUYV --png 9 $dname \
	-s brightness=${FS_BRIGHT}% -s contrast=${FS_CONT} - | \
convert - -rotate $ROTATE -channel red -threshold ${THRESH}% \
	-colorspace gray -background white \
	-alpha remove -alpha off +dither -posterize 2 -flatten - |
tee \
	>( cat > $dname ) \
	>( convert - -crop $TTE tte.png ) \
	>( convert - -crop $TRH trh.png ) \
	>( convert - -crop $ETE ete.png ) \
	>( convert - -crop $ERH erh.png ) \
	> /dev/null

running just this

Code:

#!/bin/bash

set -x

ROTATE=-1.57
TTE=68x64+147+58
TRH=56x50+150+127
ETE=44x46+117+206
ERH=46x45+193+207

FS_BRIGHT=34
FS_CONT=17
THRESH=12
dname=webcam-shot.png


fswebcam --quiet -d v4l2:/dev/video0 --no-banner \
	-F 15 -S 15 -D 1 -p YUYV --png 9 $dname \
	-s brightness=${FS_BRIGHT}% -s contrast=${FS_CONT} - | \
	convert - -rotate $ROTATE -channel red -threshold ${THRESH}% $dname

I can actually see me from my laptop webcam with the red channel having been manipulated.

using that threshold maybe forgetting to pick the channel red,green,blue leaves it whatever it leaves it then with no real bases to work with when you add your black and white -- basically white maybe the only thing it has to work with.

I am not versed in image manipulations.

but if you are going to just turn it back and white in the end then what are you doing messing with the threshold for? it seems redundant to me.

removing all of the colors then calling to have it posterized on a black and white image.

I have not tested all the way down to that, but it still seems if one has a black and white image then how is one suspect to get reds and greens and blues out of it to be posterized?

Code:

-posterize levels

reduce the image to a limited number of color levels per channel.

Very low values of levels, e.g., 2, 3, 4, have the most visible effect.

it has nothing to work with, but as I said, I am not versed in image manipulations. So i might be talking out my butt.

MOD:
messing with it still, if you use the code I tore down in the second code posted then add -posterize 2 to the last line, and change the name of the image so I can compare them. I see a definite difference between the two images.

Code:

convert - -rotate $ROTATE -channel red -threshold ${THRESH}% -posterize 2  $dname

this will give you an actual image you can make out, and I changed it to resize, cropping only gives you a little peice of the image, leaving it unreconziable to what it was.

Code:

#!/bin/bash

set -x

ROTATE=-1.57
TTE=68x64+147+58
TRH=56x50+150+127
ETE=44x46+117+206
ERH=46x45+193+207

FS_BRIGHT=34
FS_CONT=17
THRESH=3
dname=webcam-shot.png

fswebcam --quiet -d v4l2:/dev/video0 --no-banner \
	-F 15 -S 15 -D 1 -p YUYV --png 9 \
	-s brightness=${FS_BRIGHT}% -s contrast=${FS_CONT} - | 
	convert - -rotate $ROTATE -channel red -threshold ${THRESH}% \
	-posterize 2 - | \
		
tee \
	>( cat > $dname ) \
	>( convert - -resize $TTE tte.png ) \
	>( convert - -resize $TRH trh.png ) \
	>( convert - -resize $ETE ete.png ) \
	>( convert - -resize $ERH erh.png ) \
	> /dev/null

for i in {0..10}; do
	[[ -s tte.png && -s trh.png && -s ete.png && -s erh.png ]] && break
	sleep 0.33
done

#for i in tte trh ete erh; do
#	ssocr -C -c digits -d -1 -T $i.png || echo "bad value"
#done

I commented out the last one because I have no idea what that is doing, ssocr, I do not even have installed, as well.

lazardo · 01-13-2019, 02:27 AM

Quote:

I am getting just an white image, so I looked up the commands for imagemagick and added -channel played with the threshold and now I am getting some black in the images.

Everything in the original script was done to grab the face of a temp/humidity sensor which appears visually as grey and black, but is actually full color.

ssocr is an open source program that extracts segmented display values from an image, for which the success rate goes from 70% to 99.9% using the script.

Play with brightness, contrast and threshold to get the image you want, remove the posterization, colorspace and background values if not using ssocr.

If you don't need to straighten the image, remove the rotation also.

Cheers,

BW-userx · 01-13-2019, 01:02 PM

oH so this was more of a share then a what am I doing wrong question.