I live alone with my cat, and often wake up on strange angles with her sleeping on the other side of my legs than she was when I went to sleep.

I’ve been wondering what the quality of my sleep is like, and how much sharing a bed with a cat (or my partner) affects it, so I borrowed an IP camera with infrared from a friend.

This camera is old and so can’t do motion detection or write footage to a remote location (only SD card).

Below is the process by which I am recording the footage via an RTSP stream, analysing it for motion, then speeding it up so that I can watch a 2 minute summary.

Some of my notes after a few nights:

  • My cat cleans herself a lot
  • My partner grinds her teeth in her sleep
  • I might have PLMS
  • I am going to try locking my cat out for a few nights to see the difference this makes
  • Play with dvr-scan -roi to specify region of interest to just detect motion of just me (head or feet), or cat
  • Haven’t done anything with audio yet
  • Try graphing motion event timeline using timestamps outputted by dvr-scan

Example output (roughly cropped to just my cat, you don’t need to watch me sleep ya weirdo):

Prerequisites

You will need a copy of ffmpeg and dvr-scan (which you’ll also need Python for). I’m running this on Windows and have my utils in my %path%

Before you go to sleep

  1. Change working dir to a new folder for each night

  2. Start capturing RTSP stream into 5 minute long video files split using clocktime (so 1:00-1:05 etc). Each file is labeled capture-xxxxx.mp4 where xxxxx is an incrementing 5 digit number.

     ffmpeg -i rtsp://user:pass@ip:port/11 -c:v copy -c:a mp2 -ar 44100 -aq 0 -ac 2 -map 0 -f segment -segment_atclocktime 1 -segment_time 300 -segment_format mp4 "capture-%05d.mp4"

The next morning

  1. Manually delete videos at the start and end that aren’t relevant (like reading in bed, getting out of bed etc)

  2. Generate a file called scan.txt that contains the command for the next step that includes all remaining videos as parameters

     powershell "'dvr-scan', (((Get-ChildItem capture*.mp4).name) -join ' -i ') -join ' -i ' | Set-Content scan.txt"

    Example scan.txt content:

     dvr-scan -i capture-00009.mp4 -i capture-00010.mp4 -i capture-00011.mp4
  3. Open scan.txt and copy the contents into your clipboard
  4. Run dvr-scan command using the contents of your clipboard.
    This outputs one file per motion event, auto-naming all files using the filename of the first provided input file with DSME_xxxx appended on the end eg. capture-00009.DSME_0000.avi, capture-00009.DSME_0001.avi etc.
    NB. dvr-scan outputs .avi and my attempts at using -c to specify a different codec were unsuccessful
  5. Manually delete videos of motion at the start and end that aren’t relevant

  6. Output the filenames of videos of motion in the format that ffmpeg wants into a file called files.txt

     powershell "(Get-ChildItem *DSME* | ForEach-Object { 'file ''{0}''' -f $_.name }) | Set-Content files.txt"

    Example files.txt content:

     file 'capture-00009.DSME_0000.avi'
 file 'capture-00009.DSME_0001.avi'
 file 'capture-00009.DSME_0002.avi'

  7. Run ffmpeg to stitch the videos together at 20x speed (for 10x change 0.05 to 0.1)

     ffmpeg -f concat -safe 0 -i files.txt -filter:v "setpts=0.05*PTS" output.avi
  8. Watch output.avi to see how badly you sleep