high performance computing on graphics processing units: hgpu.org

Autonomous navigation systems used in Unmanned Aerial Vehicle (UAV) are mostly dependent on Global Positioning System (GPS) as a primary means of aiding Inertial Navigation Systems (INS) for accurate and reliable navigation. GPS, however, has limitations in terms of indoor availability and expected signal interference in the GPS-denied environments. The motivation of this thesis is to address the development of a low cost navigation system used onboard UAVs while maintaining accurate navigation. Motivated by the new advances in visual sensor solutions in combination with traditional navigation sensors, the proposed system is based on fusing visual measurements with INS measurements to achieve comprehensive, fast, realtime, and low cost Vision Based Navigation (VBN) system for the UAV. VBN is based on localizing set of features (with known coordinates) on the ground and finding their matches in the image taken by an imaging sensor on the UAV using a scale and rotation invariant image matching algorithm. Through Collinearity equations, object space transformation parameters are then estimated such that these matches are transformed into position information. Detailed system design and performance analysis are presented where scenarios include high dynamics of the UAV and different GPS outage are introduced. To insure fast and robust image matching algorithm, modified Speeded Up Robust Features (SURF) is introduced. The proposed algorithm is implemented on General Purpose (GP) Graphics Processing Unit (GPU) using Compute Unified Device Architecture (CUDA). Moreover, the developed algorithm is compared against the traditional least square approach, with nonlinear least squares approaches for solving the collinearity equations where large tilted aerial platform is expected, to overcome the expected nonlinearity of the mathematical model of Collinearity equations. The navigation solution is then achieved by fusing the vision measurements to the Extended Kalman Filter (EKF) as Coordinate UPdaTe (CUPT) update for the INS measurements. Performance assessments results demonstrate the enhanced performance of the proposed system against stand-alone INS solutions during a GPS signal outage.