Drone Path Planning

Unmanned Aerial Vehicles (UAV) become more and more popular today. The chairman of DJI which is one of the most important UAV producers told an online magazine that their turnover increased by a factor of four in the last years [1]. This development validates the increasing popularity of UAVs. But it is still challenging to control them and it is even more challenging to film important surface structures which are for example used for virtual model reconstructions. Therefore, the thesis presents an algorithm which is able to automatically plan the shortest drone path where all surfaces of an arbitrary mesh which the user can choose were seen. One of the main components to achieve this performance is a cost function which ensures that the drone saw most parts of a virtual model and that at the same time the travelling path is short. This function is then minimized with different optimization methods like the particle swarm and the Nelder-Mead method. The other part is a graphics processing unit-based (GPU-based) rendering technique to speed up the evaluation of the cost function. The result is a drone path which can be revisited, stored and used to reconstruct a 2D texture of a 3D surface. The findings were evaluated to demonstrate the efficiency of the path and the quality of the texture reconstruction.