Indoor and outdoor localization in an unknown environment with window detection for unmanned aerial vehicles

Abstract

In this thesis, we design and build a system for unmanned aerial vehicles (UAVs). We combine, adjust and improve existing open-source algorithms for localization tasks in an indoor and outdoor environment with window detection for transitioning between the two environments. For outdoor localization, we mainly use GPS, while for indoor localization, we use Extended Kalman Filter (EKF). However, for the transitioning area where the GPS is unreliable and the environment has too few structures for EKf, we use an opening such as a window or a door for localization with a stereo camera. We compare our technique with GPS only, EKF only, manually estimated path, and ground truth from a ZED2 camera.