Complementary Vision Based Data Fusion For Robust Positioning And Directed Flight Of An Autonomous Quadrocopter

The present paper describes an improved 4 DOF (x/y/z/yaw) vision based positioning solution for fully 6\udDOF autonomous UAVs, optimised in terms of computation and development costs as well as robustness\udand performance. The positioning system combines Fourier transform-based image registration (Fourier\udTracking) and differential optical flow computation to overcome the drawbacks of a single approach. The\udfirst method is capable of recognizing movement in four degree of freedom under variable lighting\udconditions, but suffers from low sample rate and high computational costs. Differential optical flow\udcomputation, on the other hand, enables a very high sample rate to gain control robustness. This method,\udhowever, is limited to translational movement only and performs poor in bad lighting conditions. A reliable\udpositioning system for autonomous flights with free heading is obtained by fusing both techniques.\udAlthough the vision system can measure the variable altitude during flight, infrared and ultrasonic sensors\udare used for robustness. This work is part of the AQopterI8 project, which aims to develop an autonomous\udflying quadrocopter for indoor application and makes autonomous directed flight possible. \u