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
One central object handles all communication with the remote cameras and transmits the configuration settings to each.
Configuration parameters are centrally defined in an easy-to-read file format and then handed down to the cameras.
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).
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 withrcc-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
-
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
For video conversion:
- gpac # contains MP4Box tool for video conversion
[outside world / internet]
|
|
[central machine]
|
|
[network switch]
/ | | \
| | | | <- network connection
[rpi #1] | | | e.g. ethernet cables
| | |
[rpi #2] | |
| |
[...] |
|
[rpi #n]
- Central machine, can be RPi itself (as it only holds the control object, but does no computation)
- Raspberry Pi
- Main RPi board + fast SD card (+ card reader if not available on another machine)
- RPi Camera (+lens?) (depends on your specific acquisition requirements)
- RPi power supply (RPi4 requires USB-C connector)
- Display cable (RPi4 requires mico-HDMI connector)
- Ethernet cables
- Network switch (if more than one RPi), e.g. any 1GB or faster
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
-
Set up RPi hardware
- Install Raspbian ->
NOTE: Use Raspbian Buster for now. There is no PiCamera equivalent readily available for the Raspbian Bullseye libcamera apps.
- Enable camera, GPIO interfaces, and ssh in
sudo raspi-config
options - Connect hardware:
-
Camera
-
Network cable
-
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!
-
- Install Raspbian ->
-
Install this package
- Set up python, e.g. with miniconda
- Clone this repository or use
distribute_code.sh
script (Replace hostnames for your RPi) - 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
-
DHCP server on central computer. (Description only for Ubuntu)
-
Set up static IP address on network interface that serves RPi colony via network switch, with e.g.
/etc/network/interfaces
ornetplan
-
Set up DHCP server with isc-dhcp-server
-
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
-
-
Set up python environment, e.g. with miniconda
-
Install this package
- Clone this repository
- Install with
pip install rpi_camera_colony
rcc-conductor --help
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
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
from rpi_camera_colony import read_session_data
See rpi_camera_colony.control.process_sandbox
for example use of:
from rpi_camera_colony.control.process_sandbox import ConductorAsProcess
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},
}
This software is released under the BSD 3-Clause License
-
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.
-
Video acquisition with multiple RPi synchronised by a central TTL that is recorded with the camera timestamps.
-
General package for different types of video acquisition and streaming.
-> 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
sudo rpi-update
fixes this. - Be careful, this updates the RPi firmware and might have unexpected side effects!
- 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 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
# 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
# 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"