Gadalin

Applied Computer Vision and Machine Learning

• Computer Vision Preprocessing
• Convolutional Neural Network Character Recognition
• Autonomous Driving Control

See the Team 13 - Final Report for a description of the project and outcomes.

Galadin in Action

File description of repo

• ros_ws/: Contains the ROS workspace.

  • build/: Contains the build files for ROS packages.

  • devel/: Contains the compiled binaries for ROS packages.

  • src/: Contains the source code for ROS packages.

    • my_controller: Our robot controller

      • launch/: Contains the launch file to launch the plate detection and controller nodes in parallel

      • nodes/: Python files for implementing the control algorithms

        • robot_controller : The main robot controller.
        • plate_detection.py: A seperate node for automated plate gathering.
        • vision_processing : Helper functions for machine vision across all components of the competition.
        • inner_loop_PID.py : Python class object for PID control agent in the inner loop of the course.

      • saved_images : Directory for saving snapshots to.

      • scripts/ : Bash scripts for automation of testing and verification and secondary plate gathering

      • plate_data_generation : A package from early development for gathering plates overnight at still positions

        • launch : Contains launch file for the package
        • `nodes/plate_data_snapshots.py : A script for automated capture of plates from the environment after launch
        • robots_many.launch : Launch file to replace robots.launch in the competition package for plate gathering with 8 robots
        • simulate.sh : Bash script for automation of relauching ROS overnight for plate gathering

      • 2022_competiton : The competition environment provided for the course

• cnn_trainer/: Contains the code and training data for making CNNs used in the competition.

  • cnn_alpha/ : CNN for reading plate characters.
    • placards/ : The final training data set that was used for plate character recognition.
    • wandb/ : Training data for the final models that were implemented.
    • weights/ : Saved weights from each of the training epochs of the last trained model.
    • alphachar_image_processor.ipynb: Notebook for training.
    • model.json: A trained model that is saved
    • model.h5: Saved weights and parameters to accompany model
  • cnn_parking_numbers/ : CNN for reading the parking IDs off the palcards
    • parking_image_processor.ipynb: Notebook for training.
    • model.json: A trained model that is saved
    • model.h5: Saved weights and parameters to accompany model

Useful aliases for the competition source

alias teleop='rosrun teleop_twist_keyboard teleop_twist_keyboard.py cmd_vel:=R1/cmd_vel'
alias camfeed='rosrun rqt_image_view rqt_image_view'
alias edit_source='nano ~/ros_ws/devel/setup.zsh'
alias runsim='~/ros_ws/src/2022_competition/enph353/enph353_utils/scripts/run_sim.sh -vpg'
alias robot_controller='roslaunch my_controller robot_controller.launch'
alias runsim_photomode='~/ros_ws/src/2022_competition/enph353/enph353_utils/scripts/run_sim.sh -g'
alias plate_gen='python3 ~/ros_ws/src/2022_competition/enph353/enph353_gazebo/scripts/plate_generator.py'

Here are some useful ROS commands for the competition:

• teleop: Turns on keyboard control (I turned mine off by default)
• camfeed: Opens an image topic viewing panel to view multiple topics at once
• edit_source: Quickly opens the competition source file to make changes like adding aliases
• runsim: Boots up Gazebo and runs the simulation in regular mode
• runsim_photomode: Runs the simulation without vehicles and pedestrians for working on machine vision
• robot_controller: Runs the launch file in the my_controller package
• plate_gen: Manually regenerates plates (not working through the default repo's run_sim launch)

To use these commands, you can simply type them into your terminal or add them to your .bashrc file for quick access.

Plate Detection Data Flow