PERC-Optimize-ZED-Data-Pipeline

[dinodeep]

Task Description:

Serializing and unserializing camera images can be quite expensive which is what the current pipeline setup is. Usually, to get the car started, we run

• ros2 run perceptions zed_node (create the ZEDSDK for both cameras and publish images)
• ros2 run perceptions yolov5_*_node (subscribe to images published by ZEDSDK)
  However, when running at race time, this pipeline has some unnecessary overhead with serializing and de-serializing the camera images, so this PR is going to give us the option to simply allow the yolov5_*_nodes to own ZEDSDK classes by initializing them in the DataNode and updating the camera images on a timer.

NOTE: this will prevent camera images from being published, so we would need to perform publishing of the compressed images for DAQ Live inside of the DataNode if we are still planning on using that

Task Check-list:

• Add optional parameter to DataNode to own left, right, or both cameras (or None for just using all subscribers)
• When non-None parameter passed in, initialize ZED SDK for that specific camera inside of the DataNode, and use a timer to update the images at some defined update rate
• Analyze the performance of this (do we have any issues when also running with the LiDAR pipeline?)

Task Timeline:

• Update Jira with Task Timeline

Comment any questions and tag reviewers.