This video ingestion pipeline leverages GStreamer and GStreamer Daemon (developed by RidgeRun) to control video streaming tasks on demand. The current core functionality is 1) the ability to start/stop persistent multi-file recording, 2) taking retroactive video snapshots on command, 3) taking still image snapshots from cameras, and 4) logging hardware and stream performance over time.
The setup of the video ingestion pipeline is based on configuration files. The configuration file is specified in the command line call to the pipeline and is parsed out as one of the first operations during instantiation. A sample config file is provided in './config/sample.config'; it lays out the file structure, syntax, parameters, and options. That information is also repeated in this file in Section 3.
Specification of the configuration file, as well as the other command line usage is discussed in Section 2. The config file and the destination root directory for storing the video ingestion session files are the only required options.
This repository contains only pure Python code, which does not itself require installation or setup. However, it does
have additional Python dependencies, which are included in 'requirements.txt'. They can be installed using
~/I24-video-ingest$ pip3 install requirements.txt. Note that installation of psutil may require an OS-level
dependency (see here first if errors).
There are also numerous OS-level dependencies:
In the future, a list of shell commands may be provided for installation of all dependencies.
python3 pipeline_management.py -c <config-file> -r <session-root-directory> [-t] [-m <resource-monitor-interval>] [-v] [-h]
-c/--config_file: (required for run) relative or absolute file path for session config file-r/--root_directory: (required for run) location in which to make the session directory where files are stored-t: run startup tests, which include running an image and video snapshot-m/--resource_monitor_interval: number of seconds between resource monitor logging (unspecified = monitor off)-v: print version and author information, then exit-h/--help: print usage information, then exitAt this time, a shebang (#!/usr/bin/env python3) is not included in pipeline_management.py, so the python3 command is needed.
nohup python3 pipeline_management.py [OPTIONS] &
Note that the ampersand (&) is required at the end to detach the process. This command will automatically send STDOUT and STDERR to file at ./nohup.out. To specify a log file for this output, use the following command, where output.log is your chosen destination for the log file.
nohup python3 pipeline_management.py [OPTIONS] > output.log &
Alternatively, to suppress logging this output, divert STDOUT and STDERR ("2>&1") to /dev/null.
nohup python3 pipeline_management.py [OPTIONS] > /dev/null 2>&1 &
The parent process ID (PID) will be displayed after this command (hit ENTER to continue with terminal). The parent PID can also be found in the session header file (session_directory/_SESSION_INFO.txt).
ps ax -H | grep pipeline_management.py
This will show processes hierarchically so that the parent process can be more easily identified.
kill -2 PID
Signal number 2 sends SIGINT (which is essentially a KeyboardInterrupt). SIGINT (-2) and SIGTERM (-15) are caught by main(). If a process is hanging and will not exit with -2 or -15, use SIGKILL (-9) to force.
General configuration file information is as follows. Additional parameter-specific information is below and can be found in the sample markdown file (./config/sample.config).
__CAMERA__
# name should be generic with no collisions (optional, default=camera%d)
name==camera0
# description will be written to file metadata for convenience
description==test bench camera 1
# authentication is 'user:password' format
rtsp_authentication==root:password
# address is IP:port/...
rtsp_address==192.168.0.124:554/axis-media/media.amp
# (optional) sets up logging of camera stream progress - values 'pregressreport' and 'appsink' trigger, all others don't
# note that this reporting seems to have a significant CPU impact (i.e., one full thread)
# 'progressreport' - logs a bus message generated by the progressreport element
# 'appsink' - keeps track of and logs a frame count received through appsink triggers
report==appsink
# (optional) interval (in seconds, if 'progressreport'; in frames, if 'appsink') to periodically log camera progress
report_interval==300__IMAGE-SNAPSHOT__
# Required enable declaration (use case-insensitive 'true'/'false')
enable==true__VIDEO-SNAPSHOT__
# Required enable declaration (use case-insensitive 'true'/'false')
enable==true
# (optional) Length of historical video buffer in seconds to be held for snapshot; default=60 seconds
# Not to be confused with live length of snapshot
buffer_time==30
# (optional) Default duration in seconds of video snapshot for cases when duration is not given in trigger.
default_duration==20__PERSISTENT-RECORDING__
# Required enable declaration (use case-insensitive 'true'/'false')
enable==true
# (optional) File naming template/convention for segmented recording per camera
# Filename template must contain '%d' to denote segment number, which does not roll over at max files.
# Camera name must be denoted in filename or implied directory (max once for each) using '{cam_name}' in the template
# Any directories implied in filename template will be created (including those containing camera name '{cam_name}')
# Relative directories (inside session directory) must be started with './', otherwise interpreted as absolute
# Default: './recording/record_{cam_name}_%05d.mp4'
recording_filename==./recording/record_{cam_name}_%05d.mp4
# (optional) Maximum amount of video time in minutes contained in each segment of file; default=15 minutes
segment_time==15
# (optional) Maximum number of segment files, per camera, kept in storage location (0 = no limit); default=0
maximum_segment_files==0
# (optional) Maximum estimated recording size in GB, per camera; automatically translated to number of segment files
# assumes 5.5 MB/s bitrate per camera (in parameters.py)
# this parameter overrides `maximum_segment_files`; default is not active
# maximum_camera_storage==1000This utility is designed to provide frame counts (future statistics/analytics later) for video files in a session recording directory. It finds files matching the format of the persistent recording file name found in the session configuration file ('_SESSION_CONFIG.config') and runs an FFprobe command on each to get the number of frames in the video container:
ffprobe -v error -select_streams v:0 -show_entries stream=nb_frames -of default=nokey=1:noprint_wrappers=1 file.mp4
If video recording is in progress, it is recommended to use the -d option for the frame query call, so that the last
file in the recording sequence (being actively written to) is not included (for consistency and file access conflict).
python3 query_frames.py [-h] [-p] -s <session-directory>
-s/--session_directory: (required for run) path of the session directory where files are stored-p/--print_output: flag to print output of frame counting as it is being written to file-h/--help: print usage information, then exit-d/--drop_last_file: flag to not query the last file in recording sequence, in case recording is actively occurringThere are still some lingering issues related ot logging and progress reporting that would be helpful to complete. A better way of handling interrupts from detached processes is needed. Future releases will also focus on making the management code interactive remotely, as well as from other code running on the same machine; this will be used to trigger snapshots, start/stop recording, etc.
On the horizon is the need to pipe frames directly into computer vision code, for instance across a named pipe. The GStreamer buffer objects appear to be accessible through the appsink new-sample signal handling.