Autonomous Localization and Tracking for UAVs using Kalman Filtering

Quadcopters and other drones have become more popular as they become more af- fordable and easier to use. As the accessibility increases, people discover new fields of application and one such potential use is for filming aerial footage. However, piloting a quadcopter requires a lot of training and also requires the pilot to have a visual of where the drone is. This thesis explores on how the drone can be completely automated without the need of a pilot. Using a small device called a tracker, the user simply presses a button which makes the drone autonomously take off and begin following the person carrying the tracker. The drone will keep a safe distance from the user, while always pointing its camera towards the tracker device with the help of gimbal servos to control camera angle. The quadcopter can adapt to different altitude topologies, for example if the user is climbing or skiing downhill. At any point, the user can land the drone with the click of a button. This thesis focuses on how cheap drone sensors which suffer from noise and inaccuracy, can be fused together with the help of a Kalman Filter algorithm to generate more accurate localization data for better autonomous navigation. The Kalman Filter algorithm has previously been too complex to run on quadcopter hardware, but new hardware advancements has opened the possibility to explore and research its applicability on a drone. Better sensors exist, but these are very expensive and can cost more than the drone itself. The thesis begins with describing the abilities and limitations of each sensor, then showing how they can be fused together for more accurate navigation data than what the sensors can individually produce by themselves