Stereo vision-based system for detection, track and capture of intruder flying drones

In this thesis, the design and implementation of an autonomous system that will equip a multi-rotor unmanned aerial vehicle (UAV) for visual detection and tracking of other UAVs is presented. The results from detection and tracking are used for real-time motion planning.

The goal is to effectively detect unwanted UAVs, track them and finally capture them with a net. Having a net that traps the UAVs and enables dragging intruders to another location is of great importance, since these could be carrying dangerous loads.

The project consists of three main tasks: object detection using a stereo camera, video tracking using a Kalman filter based algorithm, and lastly executing an optimal flight plan to aim a net at the detected intruder UAV. The computer vision, motion tracking and planning algorithms are implemented as ROS nodes what makes them executable on a reduced size onboard computer that is installed on the aerial vehicle.

Previous work related to this project consists of either a UAV detection system with computationally heavy algorithms or a tracking algorithm that does not include information about the dynamics of the UAVs. For the capture methods, previous ideas do not consider autonomous decisions or an optimized method to guarantee capture. In this thesis, these three aspects are considered to develop a simple solution that can be mounted on any commercially available UAV.