3D-Localization of on Unmanned Aerial Vehicle Based on Combination of Inertial Motion, GPS and Vision Information

Abstract

Unmanned aerial vehicles (UAVs) have been developed to replace human operation in complex and hazardous environments. Continuity of a UAV operation when GPS is degraded or denied is crucial in many applications such as flying near high buildings and trees, or flying outdoor- to-indoor. The state estimation is used to determine the UAV position and other states from the sensors fusion algorithm. In order to estimate position of an UAV which is a nonlinear dynamic system, accuracy, fast response and less computational burden are important requirements. In the first part, this thesis compares performance of three state estimation algorithms: inertial navigation, robot localization and Ethzasl MSF frameworks. They are implemented on ROS (Robot Operating System) to control the same UAV platform. In the experiment, GPS measurement is referred as absolute ground position for short and long flight dataset. Then, the state estimators are investigated in term of accuracy, speed and computational burden. The experimental results show that Ethzasl MSF framework has outperformed good estimation response, acceptable accuracy and reasonable computational burden. Moreover, its flexibility allows users to add other sensory information for complex scenarios. In second part, an algorithm for 3D-localization during transition between indoor and outdoor environments for a UAV is presented. Localization inputs are based on information from GPS, inertial measurement unit (IMU), monocular camera and optical flow sensor. The information is carefully selected corresponding to the operating environment regarding the GPS quality indicator which based on GPS gradient of variance. After that, the proposed smoothing offset approach is employed to smooth the position estimation. The selected sensor data are filtered by indirect extended Kalman filter for localization and extrinsic sensor calibration in real-time. The results show, the proposed smoothing offset generates a seamless and reasonable flight trajectory of UAV for indoor-outdoor transition. Moreover, the method of decision-making to cutoff GPS measurement even when it experiences poor signal quality can still outperform conventional GPS-based cutoff method in terms of response time.