A Framework for Vision-Based Multiple Target Finding and Action Using Multirotor UAVs

This paper presents a framework for vision-based target finding and action using a multirotor UAV system. The proposed framework detects and tracks a set of ground targets by using a vision-based position estimation technique. An internal map created using the relative locations of the adjacent targets is used to overcome the vision-based position estimation error and GPS noise and drift. The framework was implemented using the Robotic Operating System (ROS) and tested in the Software in the Loop (SITL) Simulation with the Gazebo robotics simulator. The test results demonstrate that the framework is robust to drift and errors, and able to perform the intended tasks successfully.