A Large Scale Inertial Aided Visual Simultaneous Localization And Mapping (SLAM) System For Small Mobile Platforms
Date of Award
Doctor of Philosophy (PhD)
In this dissertation we present a robust simultaneous mapping and localization scheme that can be deployed on a computationally limited, small unmanned aerial system. This is achieved by developing a key frame based algorithm that leverages the multiprocessing capacity of modern low power mobile processors. The novelty of the algorithm lies in the design to make it robust against rapid exploration while keeping the computational time to a minimum. A novel algorithm is developed where the time critical components of the localization and mapping system are computed in parallel utilizing the multiple cores of the processor. The algorithm uses a scale and rotation invariant state of the art binary descriptor for landmark description making it suitable for compact large scale map representation and robust tracking. This descriptor is also used in loop closure detection making the algorithm efficient by eliminating any need for separate descriptors in a Bag of Words scheme. Effectiveness of the algorithm is demonstrated by performance evaluation in indoor and large scale outdoor dataset. We demonstrate the efficiency and robustness of the algorithm by successful six degree of freedom (6 DOF) pose estimation in challenging indoor and outdoor environment. Performance of the algorithm is validated on a quadcopter with onboard computation.
Qadir, Ashraf, "A Large Scale Inertial Aided Visual Simultaneous Localization And Mapping (SLAM) System For Small Mobile Platforms" (2016). Theses and Dissertations. 2065.