The way I make the movies, is a 3 step process.
Step 1 - program runs all night, just grabbing frames from oculus, and storing them as fits files. I have the original version with the wider lens, so I set the bounds to cut off the black pieces, only grabbing the middle 1040x1040 of the sensor. This just ends me up with a directory of fits files, which your script could easily do too.
I'll get the source for this into a public location somewhere thru the day today.
Step 2 - I run a script against the directory of fits files, which produces a directory of jpeg files.
Step 3 - Use mencoder to make the movie from the jpegs.
Here is the script that does the conversions, it is set up to spawn 8 jobs in parallel (I use an 8 core machine to do it). Input filenames are expected to be in a format like this, 20150821134112.fits, where the filename is generated from the timestamp at the time it's written out, it's fairly obvious I think, yyyymmddhhmmss.fits. If one wanted to get really fancy, could probably write something to pull that info out of the fits file, but, I didn't get that fancy. the script further shaves a bit off of each image, partly because it's all trees here anyways, and partly to get the final size to something youtube is happy with. All the jpegs get dumped into a directory called test, which I clean out before each run.
#!/bin/bash
NUM=1
DIRNAME=Sept12
for NN in $(ls -rt $DIRNAME/*.fits)
do
tt=$(printf "%05.0f" $NUM)
ty=$(echo $NN | sed 's/.fits//g')
ty=$(echo $ty | sed "s/$DIRNAME//g")
ty=$(echo $ty | tr -d '/')
yr=${ty:0:4}
mo=${ty:4:2}
da=${ty:6:2}
h=${ty:8:2}
m=${ty:10:2}
s=${ty:12:2}
echo $NN $tt $h:$m:$s
#for use without timestamps
convert -shave 160x160 -rotate -90 -flop -normalize -gamma 0.8 -quality 100 $NN test/$tt.jpg &
### use for timestamps added
##convert -shave 160x160 -rotate -90 -flop -normalize -gamma 0.8 -fill white -draw "text 50,680 '$tt - $yr/$mo/$da $h:$m:$s'" -quality 100 $NN test/$tt.jpg &
NUM=$((NUM+1))
while [ $(jobs | wc -l) -ge 8 ] ; do
#echo waiting
sleep 0.10
done
done
Once you have all the jpegs, then it's easy to make the movie
mencoder mf://test/*.jpg -of lavf -lavfopts format=mp4 -ovc x264 -x264encopts bitrate=80000:global_header:threads=auto:bframes=0:crf=18 -o output.mp4 -mf fps=30
After uploading to youtube, the final output looks like this:-
Keep in mind, youtube completely destroys the quality, watching it locally the quality is MUCH higher.
Another step I recently experimented with, and it worked out. I added a callout to an external script every time my program gets a new data file, and that script just creates a jpeg and puts it onto a webserver directory. When the allsky is running, we can get a 'live view' by loading a web page from an internal web server.
