4256-18-ZED

Threading model

The Python vision code runs six threads which communicate with each other using Python's MPMC queue.Queue. Threads are encapsulated around classes, and execute their run() methods in a loop. Here is a list of the threads, their functions, and the queues they use to communicate:

• ZED()

  • Inputs: The ZED hardware
  • Outputs: zed_odometry_queue, zed_image_queue
  • Purpose: This runs in a tight loop getting data from the ZED and pushing it to the other threads.

• USB(input_camera, destination_queue) (two threads, one each for left and right cameras)

  • Inputs: the specified input camera
  • Outputs: the specified destination queue (camera_queue_L or camera_queue_R)
  • Purpose: Reads a single camera using OpenCV and posts its image to a dedicated queue in a tight loop. OpenCV reads the camera at 30Hz, so there's a lot of waiting in this loop. There are two USB threads...one for the left-facing camera and one for the right-facing one.

• Stitching()

  • Inputs: zed_image_queue, camera_queue_L, camera_queue_R
  • Outputs: stitched_image_queue
  • Purpose: Process raw ZED binocular images into a "stiched" overhead image

• NetworkTables()

  • Inputs: zed_odometry_queue
  • Outputs: NetworkTables
  • Purpose: Take the latest updates from zed_odometry_queue and post them to the global NetworkTables.

• MJPEG.ThreadedHTTPServer()

  • Inputs: stitched_image_queue
  • Outputs: supports a GET handler to send the three up-to-date images to the drive station
  • Purpose: provide up-to-date imagery to the drive station