Skip to content

Latest commit

 

History

History
39 lines (35 loc) · 2.11 KB

README.md

File metadata and controls

39 lines (35 loc) · 2.11 KB
Raw

Orion

A vision and pose estimation system for FIRST Robotics Competition robots. Inspired by Team 6328's Northstar and Team 1678's fork of it, Polaris.

Currently only tested on an Orange Pi 5 running Armbian Bookworm 24.5.5, with an Arducam OV9281 camera

Features

  • Tag detection using OpenCV's ArUco module
  • Multi-target and single-target pose estimation
  • Hardware video decoding using GStreamer
  • Configuration and output over NT4
  • MJPEG stream for debugging
  • Achieves 70-90 FPS with tested configuration

Setup

  1. OpenCV doesn't come with GStreamer support by default, so we need to build it from source. Run the following command: 
    source build-opencv.sh
    This will install OpenCV and GStreamer's dependencies and build an OpenCV Python wheel from source. This may take a while (around 45 minutes to an hour on an Orange Pi 5).
  2. Poetry is used to manage dependencies. If it's not installed, follow the instructions here, and then run the following to set up a Poetry virtual environment and install all dependencies: 
    poetry install
  3. Make a copy of ./device-config/example-network-config.json as ./device-config/network-config.json, and set the fields as needed. The device_id field will be the name of the subtable of the orion table that this device will use, and should be unique. The server_ip field should be the IP address of the NT4 server. This will usually be the RoboRIO's IP (10.TE.AM.2). The stream_port will be the port that the MJPEG stream will be served on. This only needs to be changed if there is more than one instance running on the same device.
  4. Use CalibDB to calibrate your camera and export the calibration file using OpenCV formatting. Save this file to ./device-config/calibration.json.
  5. Run 
    poetry run python -m orion
    to start the vision system.