Object Contextualization with Monocular Camera for a UAV
This package is designed to work hand in hand with YOLO and Ardupilot to give the UAV awareness of where detected objects are in the UAV's navigational frame. This package uses the intrinsics of the camera (the field of view) as well as the mounting angle and the Drone's position.
Why use this package?
In order to get the position of a detection the traditional solution is to use a stereo camera. However, stereo cameras have 2 significant drawbacks.
1.) They are much hevier than monocular cameras
2.) Stereo algorithms require more computational processing
By using this package your UAV will have more available computational power and be physically much lighter (This can create significant flight time increases)
What are the limitations of this package?
This package make 2 key assumptions
1.) Detected objects reside on the ground
2.) Detected object's height are much less than the altitude of the drone
Original Use Case
This package was orginally used to detect roombas for IARC's mission 7. The position of all the roombas fed back to a state machine running onboard the drone which then made decisions on which roombas to pursue.
https:/
Information Flow Chart
Dependencies
- darknet_ros
- iq_control_node
cd ~/catkin_ws/src git clone https://github.com/Texas-Aerial-Robotics/darknet_ros.git
Installaton
cd ~/catkin_ws/src git clone https://github.com/Texas-Aerial-Robotics/Transformations-Ros.git
Build
catkin build
Running package
To test your system use the gazebo + ardupilot SITL simulation
TODO make general instructions for running gazebo + ardupilot SITL simulation
Run YOLO roslaunch darknet_ros darknet_ros.launch
Run Transformation node roslaunch transformations_ros transformations.launch