Changing the input streamer or file path to ros2 node publisher

[AlexKnish]

Hey all, and thank you for this incredible project!
I would like to know what are the changes that need to be done in order to convert the input of the node from camera/mp4 file path to a topic of a node that publisher bag file of frames.
which class should be changed or what kind of topic should I use? or other modifications I should take in consider.
best regrades,
AK

[AlexKnish]

Abbreviation of the question above, what should i change in the deepstream pipeline flow, in order to replace the input stream with a ros topic subscriber which receives frame by frame images (using Image msg converted by cv2 in python
cv_image = self.bridge.imgmsg_to_cv2(msg)

[AsawareeBhide]

Hi! This project currently supports video input through webcam/file.

Converting ROS to OpenCV images will not give you the latency optimizations that the underlying DeepStream pipeline provides. Could you please tell us what you want to achieve using CvBridge?

If interested in accessing image data, you could try using deepstream-imagedata-multistream.

You could also try using RTSP input with your DeepStream application; refer to deepstream-rtsp-in-rtsp-out.

[AlexKnish]

Thanks @AsawareeBhide for your response!

1. Short overview, the system image analysis process, is inside ros2 node and receives frames by DDS messages, that sent from a mDVR component, than using self.bridge.imgmsg_to_cv2(msg) the image topic is converted to cv2 and continue to YOLO NN for inference.
2. I wish to replace the analytic process with the deepstream pipeline.
3. In order to support the existing flow (using messages), I wrapped the Deepstream with ROS2 subscriber (with this project).
4. But didnt find a plugin that supports the receives or transformation from ROS\DDS topics to some thing that supported by deepstream input.

So what should be a good practice to convert DDS/ROS messages to a deepstream input?

[AlexKnish]

Hey there,
Still intrigued to find a solution to this problem, so I really appreciate any feedback/ help.

[doutdex]

Any approach? Very interesting to move some ai to models to ROS using real-time cameras on board robot