TLDR: Publish frames from a connected camera (e.g. USB webcam, or alternatively an MJPEG/RTSP stream) to an MQTT topic. The camera stream can be viewed in a browser with Streamlit or Home Assistant. Configuration is via config.yml
Long introduction: A typical task in IOT/science is that you have a camera connected to one computer and you want to view the camera feed on a different computer, and maybe preprocess the images before saving them to disk. I have always found this to be may more work than expected. In particular working with camera streams can get quite complicated, and may lead you to experiment with tools like Gstreamer and ffmpeg that have a steep learning curve. In contrast, working with MQTT is very straightforward and is also probably familiar to anyone with an interest in IOT. Whilst MQTT is rarely used for sending files, I have not encountered any issues doing this.
mqtt-camera-streamer
uses MQTT to send frames from a camera connected to a computer over a network at low frames-per-second (FPS). A viewer is provided for viewing the camera stream on any computer on the network. Frames can be saved to disk for further processing. Also it is possible to setup an image processing pipeline by linking MQTT topics together, using an on_message(topic)
to do some processing and send the processed image downstream on another topic.
Note that this is not a high FPS solution, and in practice I achieve around 1 FPS which is practical for tasks such as preprocessing (cropping, rotating) images prior to viewing them. This code is written for simplicity and ease of use, not high performance.
Use a venv to isolate your environment, and install the required dependencies:
$ (base) python3 -m venv venv
$ (base) source venv/bin/activate
$ (venv) pip3 install -r requirements.txt
Do not use a venv but install openCV system wide using:
$ sudo apt install python3-opencv
$ pip3 install -r requirements.txt
I have not tested Streamlit on the Raspberry pi, but you can use the viewer on another machine (WIndows, OSX) so don't worry.
The check-cameras.py
script assists in discovering which cameras are on your computer. If your laptop has a built-in webcam this will generally be listed as VIDEO_SOURCE = 0
. If you plug in an external USB webcam this takes precedence over the built-in webcam, with the external camera becoming VIDEO_SOURCE = 0
and the built-in webcam becoming VIDEO_SOURCE = 1
.
To check which cameras are detected run:
$ (venv) python3 scripts/check-cameras.py
You then configure the desired camera as e.g. video_source: 0
. Alternatively you can configure the video source as an MJPEG or RTSP stream. For example in config.yml
you would configure video_source: "rtsp://admin:password@192.168.1.94:554/11"
Use the config.yml
file in config
directory to configure your system (mqtt broker IP etc) and validate the config can be loaded by running:
$ (venv) python3 scripts/validate-config.py
Note that this script does not check the accuracy of any of the values in config.yml
, just that the file path is correct and the file structure is OK.
By default scripts/camera.py
will look for the config file at ./config/config.yml
but an alternative path can be specified using the environment variable MQTT_CAMERA_CONFIG
To publish camera frames over MQTT:
$ (venv) python3 scripts/camera.py
To view the camera stream with Streamlit:
$ (venv) streamlit run scripts/viewer.py
Note: if Streamlit becomes unresponsive, ctrl-z
to pause Streamlit then kill -9 %%
. Also note that the viewer can be run on any machine on your network.
To save frames to disk:
$ (venv) python3 scripts/save-captures.py
To process a camera stream (the example rotates the image):
$ (venv) python3 scripts/processing.py
You can view the camera feed using Home Assistant and configuring an MQTT camera. Add to your configuration.yaml
:
camera:
- platform: mqtt
topic: homie/mac_webcam/capture
name: mqtt_camera
- platform: mqtt
topic: homie/mac_webcam/capture/rotated
name: mqtt_camera_rotated
Need an MQTT broker? If you have Docker installed I recommend eclipse-mosquitto. A basic broker can be run with:
docker run -p 1883:1883 -d eclipse-mosquitto
Note that I have structured the MQTT topics following the homie MQTT convention, linked in the references. This is not necessary but is best practice IMO.