Skip to content

Central control for video acquisition with (many) Raspberry Pi cameras

License

Notifications You must be signed in to change notification settings

larsrollik/rpi_camera_colony

DOI Website PyPI Wheel Code style: black

RPi Camera Colony (RCC)

Central control for video acquisition with (many) Raspberry Pi cameras

Record videos in parallel with one or more remote-controlled Raspberry Pi (RPi) cameras. 🎥

A single configuration file and a few lines of code allow specific and reproducible acquisition settings for groups of cameras.

Example use with Python:

import time
from rpi_camera_colony.control.conductor import Conductor

conductor = Conductor(settings_file="configuration_file")  # Manages remote RPi
conductor.start_acquisition()  # Starts recording on all remotes

time.sleep(20)  # do nothing or something else in between

conductor.stop_acquisition()  # Stops recording on all remotes

or on the commandline:

rcc-conductor  --config-data-file CONFIG_DATA_FILE  --acquisition-name ACQUISITION_NAME

Features

A centralised control object

One central object handles all communication with the remote cameras and transmits the configuration settings to each.

A single configuration file to define reproducible multi-camera acquisition

Configuration parameters are centrally defined in an easy-to-read file format and then handed down to the cameras.

Flexible entrypoints

Multiple entrypoints for use in python scripts as well as in a single line on the commandline Additionally, all levels are directly accessible: central Conductor, remote control handlers, and on the RPi the acquisition control (see below for details).

NEW: Network video stream

Add an additional output via network video stream directly from the main config or via commandline arguments when calling rcc-acquisition.

  • config example to use with rcc-conductor as usual:
# ...

[controllers]
    [[camera_red_60]]
        description = "back view"
        address = "192.168.0.22"

        # Network stream setup:
        stream_video = True
        stream_address = "192.168.0.22"
        stream_port = 8001

# ...
  • Command-line entrypoint example:

    • Options:
    -s, --stream-video
    -sip STREAM_IP, --stream-ip STREAM_IP
                          IP address for video stream. (default: 192.168.100.31)
    -sport STREAM_PORT, --stream-port STREAM_PORT
                          Stream port (default: 8001)
    • Example call:
    rcc-acquisition --auto-start --stream-video --stream-ip 192.168.100.31 --stream-port 9898

Installation

Python dependencies

  • python >= 3.6

  • pyzmq

  • configobj

  • tqdm

  • numpy

  • pandas

    Note: Use conda to install numpy/pandas to get pre-compiled packages (See below for instructions)

On RPi only:

  • picamera
  • RPi.GPIO

Other useful packages

For video conversion:

  • gpac # contains MP4Box tool for video conversion

Example hardware architecture

  [outside world / internet]
              |
              |
      [central machine]
              |
              |
      [network switch]
      /   |     |    \
     |    |     |     | <- network connection
[rpi #1]  |     |     |     e.g. ethernet cables
          |     |     |
     [rpi #2]   |     |
                |     |
              [...]   |
                      |
                 [rpi #n]

Minimal hardware requirements

  • Central machine, can be RPi itself (as it only holds the control object, but does no computation)
  • Raspberry Pi
    1. Main RPi board + fast SD card (+ card reader if not available on another machine)
    2. RPi Camera (+lens?) (depends on your specific acquisition requirements)
    3. RPi power supply (RPi4 requires USB-C connector)
    4. Display cable (RPi4 requires mico-HDMI connector)
  • Ethernet cables
  • Network switch (if more than one RPi), e.g. any 1GB or faster

Mapping between this package & hardware

One Conductor to instruct all RPi cameras via network communication between the RemoteAcquisitionControl and PiAcquisitionControl.


    Hardware            <-->        Software


    [central machine]   <-->        Conductor
            |                           |
            |                       RemoteAcquisitionControl
            |                           |
           ...                         ...
            |                           |
       [rpi #n]         <-->        PiAcquisitionControl
                                        |
                                    Camera

Raspberry Pi setup

  1. Set up RPi hardware

    1. Install Raspbian -> NOTE: Use Raspbian Buster for now. There is no PiCamera equivalent readily available for the Raspbian Bullseye libcamera apps.
    2. Enable camera, GPIO interfaces, and ssh in sudo raspi-config options
    3. Connect hardware:
      1. Camera

      2. Network cable

      3. GPIO pin connection for TTL in/out (See pinout.xyz for board mode pins to use)

          Note: adjust pin numbers used in configuration file. Default are pin #8 for frame TTL outputs and #16 for inputs. Choose any free ground pins!
        
  2. Install this package

    1. Set up python, e.g. with miniconda
    2. Clone this repository or use distribute_code.sh script (Replace hostnames for your RPi)
    3. Install a. From Pypi
      pip install rpi_camera_colony[rpi]  # <- Note: `[rpi]` argument adds specific requirements for acquisition on RPi, but is not required for controller

    b. From Github shell pip install https://github.com/larsrollik/rpi_camera_colony[rpi] # <- Note: `[rpi]` argument adds specific requirements for acquisition on RPi, but is not required for controller

Central control machine setup

  1. DHCP server on central computer. (Description only for Ubuntu)

    1. Set up static IP address on network interface that serves RPi colony via network switch, with e.g. /etc/network/interfaces or netplan

    2. Set up DHCP server with isc-dhcp-server

    3. Set up SSH keys to allow interaction with RPi without password (otherwise cannot drop remote process!)

       ssh-keygen  # into standard file if not exists, no passphrase
       ssh-copy-id -i ~/.ssh/id_rsa HOST  # where HOST = RPi host name
      
  2. Set up python environment, e.g. with miniconda

  3. Install this package

    1. Clone this repository
    2. Install with
      pip install rpi_camera_colony

Entrypoints & levels

Easy access to central Conductor

rcc-conductor --help

Use acquisition directly on RPi

from rpi_camera_colony.acquisition.acquisition_control import PiAcquisitionControl

or

python rpi_camera_colony/acquisition --help
# or
python -m rpi_camera_colony.acquisition --help
# or
rcc-acquisition --help

One-to-one mapping of local control to remote acquisition

from rpi_camera_colony.acquisition.remote_control import RemoteAcquisitionControl

or

python rpi_camera_colony/acquisition --help
# or
python -m rpi_camera_colony.acquisition.remote_control --help

Read acquisition metadata & check for video files

from rpi_camera_colony import read_session_data

Sandbox Conductor object in separate process (python multiprocessing)

See rpi_camera_colony.control.process_sandbox for example use of:

from rpi_camera_colony.control.process_sandbox import ConductorAsProcess

Citation

Rollik, Lars B. (2021). RPi Camera Colony: Central control for video acquisition with (many) Raspberry Pi cameras. doi: 10.5281/zenodo.6414747.

BibTeX

@misc{rollik2021rpi,
    author       = {Lars B. Rollik},
    title        = {{RPi Camera Colony: Central control for video acquisition with (many) Raspberry Pi cameras}},
    year         = {2021},
    month        = jun,
    publisher    = {Zenodo},
    url          = {https://doi.org/10.5281/zenodo.6414747},
    doi          = {10.5281/zenodo.6414747},
  }

License

This software is released under the BSD 3-Clause License

Related projects with similar architectures

  • Arne Meyer's RPiCameraPlugin for the OpenEphys GUI

    Specific API for one-to-one control mappings between OpenEphys GUI plugin instances and remote RPi cameras. Inspiration for use of ØMQ communication and camera TTL integration in encoder class.

  • Deshmukh lab's PicameraPaper

    Video acquisition with multiple RPi synchronised by a central TTL that is recorded with the camera timestamps.

  • Vidgear

    General package for different types of video acquisition and streaming.

Configuration file specification

-> Note: additional picamera attributes can be used, but not all types implemented. Check below.

-> Note: acquisition_group is not specified by default, but if acquisition_name contains __ double underscores, then the acquisition_group will get auto-populated from the first segment, when split on the __. This is to create an acquisition folder organisation like: /path_to_data/acquisition_group/acquisition_name/[files]

[general]
    acquisition_name = string(default="_test_rcc_name_config")    # base name for recording
    acquisition_time = string(default="dummy_time")
    acquisition_group = string(default="")
    remote_data_path = string(default="/home/pi/data/")             # where to store all recordings on RPi
    rpi_username = string(default="pi")
    remote_python_interpreter = string(default="/home/pi/miniconda3/envs/py36/bin/python")      # path to python
    remote_python_entrypoint = string(default="rpi_camera_colony.acquisition")      # path to __main__ entrypoint
    max_acquisition_time = integer(0, 7200, default=7200)           # seconds, shut down acquisition after expiration
    save_data = boolean(default=True)  # if False, then doesn't write files on RPi
    general_setting_has_priority = boolean(default=True)  # If False, does not patch in general settings
    general_settings_to_patch_into_controller = string_list(default=list("save_data", "acquisition_time", "acquisition_group"))  # Add variables here for patching into controllers

[log]
    address = string(max=15, default="192.168.100.10")
    port = integer(default=55555)
    level = string(default="DEBUG")
    log_to_console = boolean(default=True)
    log_to_file = boolean(default=True)
    log_file = string(default="/tmp/rpi_camera_colony__logging")

[control]
    address = string(default="192.168.100.10")
    port = integer(default=54545)

[controllers]
    [[__many__]]
        description = string(default="")
        address = string(max=15, default="")
        save_data = boolean(default=True)
        ttl_channel_external = integer(default=-1)  # metadata info if recording output TTL on specific channel of other acquisition system
        ttl_in_pin = integer(default=16)
        ttl_out_pin = integer(default=8)
        ttl_out_duration = float(default=.001)

        # See for list of ALL parameters https://picamera.readthedocs.io/en/latest/api_camera.html
        framerate = integer(min=1, max=90, default=90)
        resolution = int_list(default=list(640, 480))
        vflip = boolean(default=False)
        hflip = boolean(default=False)

        brightness = integer(min=0, max=100, default=50)
        # color_effects: (128, 128) == black and white acquisition. Default is None.
        color_effects = int_list(default=list(128, 128))
        contrast = integer(min=-100, max=100, default=0)
        image_denoise = boolean(default=True)
        iso = integer(min=0, max=1600, default=0)
        led = boolean(default=False)
        preview_alpha = integer(min=0, max=255, default=255)  # DEPRECATED
        saturation = integer(min=-100, max=100, default=0)
        sharpness = integer(min=-100, max=100, default=0)
        still_stats = boolean(default=False)
        video_denoise = boolean(default=True)
        video_stabilization = boolean(default=False)
        # zoom: (x, y, w, h)
        zoom = float_list(default=list(0.0, 0.0, 1.0, 1.0))

        # Other picamera attributes / not implemented / not tested, but might work
        # awb_gains
        # awb_mode = option(default="auto")
        # drc_strength
        # exposure_compensation
        # exposure_mode
        # exposure_speed = 0
        # flash_mode = option("off", "auto", "on", "redeye", "fillin", "torch", default="off")
        # framerate_delta  # new in 1.11
        # framerate_range  # new in 1.13
        # image_effect = "none"
        # image_effect_params  # https://picamera.readthedocs.io/en/release-1.13/api_camera.html#picamera.PiCamera.image_effect_params
        # meter_mode = option("average", "spot", "backlit", "matrix")
        # rotation = option(0, 90, 180, 270)
        # sensor_mode = integer(default=0)
        # shutter_speed  # microseconds

Specific install hints & solutions

HQ camera for RPi cannot acquire at resolutions or framerates outlined in the technical description

  • sudo rpi-update fixes this. - Be careful, this updates the RPi firmware and might have unexpected side effects!

There is more than one TTL-in detected for long (>50ms) up/down states of the TTL source

  • add a bouncetime parameter to the event detection in camera.py#L196
  • remove surplus TTL-in post-hoc based on the known minimum spacing of TTL

Ip forwarding and routing on central machine

# IP forward
sysctl -w net.ipv4.ip_forward=1
# check with
cat /proc/sys/net/ipv4/ip_forward

# Package routing
# - outside interface (dhcp): enp7s0
# - inside interface (static): enp8s0
iptables -A FORWARD -i enp8s0 -o enp7s0 -j ACCEPT
iptables -A FORWARD -i enp7s0 -o enp8s0 -m state --state ESTABLISHED,RELATED -j ACCEPT
iptables -t nat -A POSTROUTING -o enp7s0 -j MASQUERADE

Update time for ssl certificates

# Check with
timedatectl status

# force update with NTP
sudo service ntp stop
sudo ntpd -gq
sudo service ntp start

# enable permanent updates
sudo systemctl restart systemd-timesyncd

Install miniconda on RPi

# Installing miniconda on RPi
wget http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-armv7l.sh
sudo md5sum Miniconda3-latest-Linux-armv7l.sh # (optional) check md5
bash Miniconda3-latest-Linux-armv7l.sh # -> default directory should be: /home/pi/miniconda3

# Add conda to path
echo 'export PATH="/home/pi/miniconda3/bin:$PATH"' >> .bashrc
source .bashrc # or re-connect

# Create conda environment and install basic packages (e.g. dependencies for this package)
conda config --add channels rpi
conda create -y -n py36 python=3.6 numpy pandas pyzmq

echo 'source activate py36' >> .bashrc
source .bashrc # or re-connect

# Re/install RCC with `[rpi]` option to install picamera and GPIO packages on RPi.
pip uninstall rpi_camera_colony -y
pip install --upgrade rpi_camera_colony[rpi]

Version: "0.5.0"