Vision Aided Navigation of a Quad-Rotor for Autonomous Wind-Farm Inspection

This work presents a vision based navigation system for an autonomous drone that is capable of recognizing and locating wind mills. WindMillNet, a Deep Neural Network created for this purpose was specially trained to recognize wind mills on camera images using transfer learning techniques. The drone powered by WindMillNet scans the horizon to find wind mills, and after perceiving a wind mill, navigates towards it, with the goal of performing its inspection. A hierarchical control system, implemented in the drone provides stability and control of its movements. Our framework was designed using a cyber-physical systems approach using high-level abstractions in modeling, communication, control and computation.