chipturner/rtsp-timelapse-scripts

Some scripts I made for creating a timelapse from an RTSP stream

Overview

These scripts are basically (slightly) cleaned up versions of personal scripts I used to turn an RTSP stream into a timelapse video. I capture video to a NAS via UniFi cameras with grab-timelapse-frame.py via cron and periodically use ffmpeg with the files provided by filter-timelapse-frames.py to produce an actual video.

Interesting, Motivating Details

My project was backyard construction. This means I am only interested in weekday work when there is daylight, hence defaults that are tuned to that (e.g. using astral to determine when sunrise and sunset are to avoid capturing frames, and skipping weekends).

I want to capture roughly one image every 10 seconds so I can have a smooth video per day but also can then sample over longer periods. Hence:

• grab-timelapse-frame.py passes filename and directory names to strftime so as to avoid too many files in a single directory.
• grab-timelapse-frame.py is made robust by intentionally being short lived and run from cron; no complexities from systemd though you do lose some monitoring/management. Oh well, it works for me and seems worth it.
• filter-timelapse-frames.py supports a --sample parameter to only print every Nth matching file for when you want to produce faster videos by not including every frame. (TODO: eventually process images directly with ML to filter in/out people, animals, bitcoins, etc).
• Sunrise and sunset are relative to wherever you are, so grab-timelapse-frame.py accepts a --city parameter (full list of cities is provided by astral's geocoder.py module.

Basic Usage

First, install ffmpeg with whatever libraries are most approriate for output formats you want. I like x264 for speed and compression ratio.

Capturing

First, you need to capture frames. grab-timelapse-frame.py is designed to be run in a cron. It is a short-running script that, ideally, you invoke every minute and let it run for one minute. During that time it will produce N evenly spaced out frames into the specified output directory. Example cronjob:

* * * * *	/path/to/grab-timelapse-frame.py --output-directory /path/to/outputs --output-filenames cam1 --url rtsp://HOST:PORT/PATH

NOTE: If you use strftime in the output strings, don't forget to backslash % in your crontab -- cron translates them to newlines..

Creating a video

Congratulations, you have a pile of png files. To make a video, you need to use ffmpeg with a list of files. Since you likely have too many files, you want to filter:

$ ./filter-timelapse-frames.py /path/to/pile --sample 10 > /tmp/filelist
$ ffmpeg -r 30 -f concat -safe 0 -i <(sed 's/^/file /' /tmp/filelist) -c:v libx264rgb -preset veryslow -crf 21 -vf fps=30 /path/to/output.mp4

You can google for what the ffmpeg line does and learn how to produce other output file formats, but the above works well for me. Note it will be slow. The <(sed ...) bit is to prefix each line with 'file ' to produce an instruction file that ffmpeg likes.

And you're done! Enjoy your fun video. VLC is probably the best tool to view it in.

Reducing PNG Storage Space

The PNG files created by ffmpeg are not as optimized as they could be. ImageMagick can help with this, but if you end up with a large pile of files, reprocessing them all can be annoying. The generic process-file-once.py script here is meant to be used avoid re-running expensive operations on files by tracking if a given command line has been run on a file before and, if so, skipping running. This makes commands idempotent no-ops and lets you lazily re-run find commands over and over without wasting time reprocessing already processed files.

Here's how I use it in this project:

$ find /path/to/pile -name '*.png' | parallel --progress -I{} 'process-file-once.py {} mogrify -format png -define png:compression-level=9 {}'

The use of GNU parallel here effectively saturates your CPU.

This saves me roughly 25% space.