|
|
Neural Networks in Inertial Navigation Systems
Tejmlová, Lenka ; Ochodnický,, Ján (oponent) ; Masopust, Jiří (oponent) ; Šebesta, Jiří (vedoucí práce)
The dissertation is focused on inertial navigation systems and dead reckoning positioning. The issue in the problematics is that the dead reckoning systems and inertial navigation systems are inaccurate for medium-term and long-term application due to cumulative errors, assuming that the positioning is not supported by another external system. The dissertation shows possible approaches to the issue of more accurate positioning system based only on the inertial sensors. Basically we are talking about 9-DOF inertial measurement unit that allows sensing the global acceleration, rotation rate and magnetic field strength in three particular axes. The new approach brings artificial neural networks into data processing, where proper neural network is able to recognize the character of motion leading to improvement in positioning. The description of the proposed method includes an analytical procedure of its development and, if possible, the analytical performance assessment. Proposed artificial neural networks are modelled in MATLABTM and they are used for the determination of the state of the inertial unit. Due to this determination, the position of the inertial measurement unit is evaluated with higher accuracy. An application using Qt framework was developed to create an evaluation system with user interface for standard inertial measurement unit. The designed system based on artificial neural networks was verified by experiments using real sensor data.
|
|
|
Neural Networks in Inertial Navigation Systems
Tejmlová, Lenka ; Ochodnický,, Ján (oponent) ; Masopust, Jiří (oponent) ; Šebesta, Jiří (vedoucí práce)
The dissertation is focused on inertial navigation systems and dead reckoning positioning. The issue in the problematics is that the dead reckoning systems and inertial navigation systems are inaccurate for medium-term and long-term application due to cumulative errors, assuming that the positioning is not supported by another external system. The dissertation shows possible approaches to the issue of more accurate positioning system based only on the inertial sensors. Basically we are talking about 9-DOF inertial measurement unit that allows sensing the global acceleration, rotation rate and magnetic field strength in three particular axes. The new approach brings artificial neural networks into data processing, where proper neural network is able to recognize the character of motion leading to improvement in positioning. The description of the proposed method includes an analytical procedure of its development and, if possible, the analytical performance assessment. Proposed artificial neural networks are modelled in MATLABTM and they are used for the determination of the state of the inertial unit. Due to this determination, the position of the inertial measurement unit is evaluated with higher accuracy. An application using Qt framework was developed to create an evaluation system with user interface for standard inertial measurement unit. The designed system based on artificial neural networks was verified by experiments using real sensor data.
|
host ::
RSS.