Dissertation

Digital Processing of Magnetic, Angular-Rate and Gravity Signals for Human-Computer Interaction

Ratchatanantakit, Neeranut

2021

Overview

This dissertation pursued the definition and evaluation of a processing approach for robust real-time orientation estimation of a miniature Magnetic, Angular-Rate, Gravity (MARG) module for use in a human-computer interaction system that also uses a 3-camera IR-video module for position estimation. The proposed algorithm introduces the novel idea of spatially mapping the level of trustworthiness of the magnetometer-based potential corrections to the orientation estimate. This trustworthiness level is used to reduce the strength of magnetometer-based corrections of the orientation estimate where the magnetic field distortion invalidates the assumptions necessary for those corrections. The new algorithm addresses the three research questions posed in this dissertation by 1.) Compensating for the gyroscope drift error 2.) Creating a voxel map with the values of magnetic distortion in specific regions of the operating space of the system, and 3.) Combining two different types of data to accurately track hand motion through adaptive quaternion interpolation. The algorithm was evaluated in an experiment with thirty human subjects, processing signals from one MARG module and a 3-camera IR video system. The results verified that the new algorithm, using the Gravity Vector and the Magnetic North vector with Double SLERP interpolation (GMV-D), reduced the drift of the orientation estimates in areas with and without magnetic distortion. The Kruskal-Wallis test, with the error in the Phi, Theta, and Psi Euler angles as dependent variables, was used to study 3 orientation estimation methods: Kalman Filtering, GMV-D, and its precursor, GMV-S (which uses a single SLERP operation). In the magnetically undistorted area, there were no significant differences for the Phi and Theta angles. However, in the magnetically distorted area, significant differences in method performance were found for all three Euler angles, with GMV-D consistently reporting the lowest mean rank and the Kalman Filter reporting the highest. The proposed GMV-D method makes the MARG orientation estimation more robust by fully taking advantage of the MARG operating conditions in a typical human-computer interaction application and by comprehensively utilizing all the sensing modalities available in the MARG module.

• Publisher: ProQuest Dissertations & Theses
• Subject: Biomedical engineering / Computer Engineering / Computer science / Electrical engineering
• ISBN: 9798277429938