openpose-drone-control

Drone control scheme using OpenPose and neural net classification.

Setup

1. Install OpenPose.
2. Clone this repository to the same folder that contains OpenPose.
3. Install ROS, MAVROS, Gazebo, and ArduCopter. Make sure to source /opt/ros/melodic/setup.bash and export ardupilot/Tools/autotest in order to use ROS and ArduCopter commands.

Launch

To launch, run the following commands in separate terminals:

sim_vehicle.py -v ArduCopter -f gazebo-iris --map --console
gazebo --verbose obstacle.world
roslaunch mavros apm.launch
python3 pose.py

To retrain the classification model, do the following:

1. Open build_dataset.py and edit the line pose = ... to the pose name that you want to build a new dataset for. (note: valid pose names include ['forward', 'back', 'turnleft', 'turnright', 'up', 'down', 'left', 'right', 'neutral'])
2. Run python3 build_dataset.py, then hold that pose until the script ends.
3. Run python3 train_model.py.