matlock08/deepstream-python

Sample application running in docker-compose to execute deepstream and read, publish messages from/to mqtt

Deepstream 6.0.1 with Python on Docker Compose

This project uses deepstream 6.0.1 running on a docker compose environment with mqtt from mosquitto the project is developed in python and reads messages from mqtt to pause, play or stop the pipeline as well as to disable or enable inference.

To run this you need an Ubuntu 18.04, as well as an Nvidia CUDA 11.1 capable card.

Also support to add probe callback to get access to images in pipeline or to send messages

                   ┌ queue - valve - nvinfer -nvtracker - nvinfer - nvinfer -nvinfer ┐ 
                   |                                                                 |
src - streammux - tee                                                              funnel - nvvideoconvert - capsfilter - nvmultistreamtiler - nvvideoconvert - nvosd - nveglglessink
                   |                                                                 |
                   └ queue - valve --------------------------------------------------┘

Run project

To run the project you need to create a deepstream.env file with below keys.

STREAMS is an array of sources rtsp or files

MQTT_BROKER=mqtt
MQTT_PORT=1883
MQTT_TOPIC_COMMAND=/deepstream/command
MQTT_TOPIC_MESSAGES=/deepstream/message
STREAMS="rtsp://url:port/ rtsp://url:port"
SHOW_ON_SCREEN=1
PYTHONUNBUFFERED=1

Then you can build the images by running

docker-compose build

And finally run it by

xhost +
docker-compose up

Once the two instances are running you can connect to the mqtt server on localhost 1883 with any mqtt client and send commands

topics	message	description
/deepstream/command	{"command": "play"}	Sets play status to pipeline (Default state)
/deepstream/command	{"command": "pause"}	Sets pause status to pipeline
/deepstream/command	{"command": "stop"}	Sets stop status to pipeline
/deepstream/command	{"inference": "enabled"}	Enables the inference flow (Default state)
/deepstream/command	{"inference": "disable"}	Disable the inference flow still image visible on sync