| Due to the complementary error characteristics of the Global Positioning System (GPS) and Inertial Navigation System (INS), their integration has become a core positioning component, providing high-accuracy direct sensor georeferencing for multi-sensor mobile mapping systems. Despite significant progress over the last decade, there is still a room for improvements of the georeferencing performance using specialized algorithmic approaches. The techniques considered in this dissertation include: (1) improved single-epoch GPS positioning method supporting network mode, as compared to the traditional real-time kinematic techniques using on-the-fly ambiguity resolution in a single-baseline mode; (2) customized random error modeling of inertial sensors; (3) wavelet-based signal denoising, specially for low-accuracy high-noise Micro-Electro-Mechanical Systems (MEMS) inertial sensors; (4) nonlinear filters, namely the Unscented Kalman Filter (UKF) and the Particle Filter (PF), proposed as alternatives to the commonly used traditional Extended Kalman Filter (EKF).; The network-based single-epoch positioning technique offers a better way to calibrate the inertial sensor, and then to achieve a fast, reliable and accurate navigation solution. Such an implementation provides a centimeter-level positioning accuracy independently on the baseline length. The advanced sensor error identification using the Allan Variance and Power Spectral Density (PSD) methods, combined with a wavelet-based signal de-noising technique, assures reliable and better description of the error characteristics, customized for each inertial sensor. These, in turn, lead to a more reliable and consistent position and orientation accuracy, even for the low-cost inertial sensors. With the aid of the wavelet de-noising technique and the customized error model, around 30 percent positioning accuracy improvement can be found, as compared to the solution using raw inertial measurements with the default manufacturer's error models. The alternative filters, UKF and PF, provide more advanced data fusion techniques and allow the tolerance of larger initial alignment errors. They handle the unknown nonlinear dynamics better, in comparison to EKF, resulting in a more reliable and accurate integrated system. For the high-end inertial sensors, they provide only a slightly better performance in terms of the tolerance to the losses of GPS lock and orientation convergence speed, whereas the performance improvements are more pronounced for the low-cost inertial sensors. |
