Nonlinear Estimation Techniques for High-Resolution Indoor Positioning Systems
MetadataShow full item record
The Global Positioning System (GPS) is the most popular positioning system among some operational Global Navigation Satellite Systems (GNSS). However, GNSS suffer from accuracy deterioration and interruption of services in dense urban areas and are almost unavailable indoors. Although high-sensitivity receivers improve signal acquisition indoors, multipath is still be a challenging problem that affects accuracy especially indoors where a direct line of sight between transmitter and receiver almost never exist. Moreover, the wireless signal features are significantly jeopardized by obstacles and constructions indoors. To address these challenges, this research came in the context of proposing an alternative positioning system that is designed for GPS-denied environment and especially for indoors. Cramer-Rao Lower-Bound (CRLB) analysis was used to estimate the lower bound accuracy of different positioning methods indoors. Based on CRLB analysis, this research approached the wireless positioning problem indoors utilizing received signal strength (RSS) to achieve the following: 1) Developing new estimation methods to model the wireless RSS patterns in indoors. 2) Designing adaptive RSS-based wireless positioning methods for indoors. 3) Establishing a consistent framework for indoor wireless positioning systems. 4) Developing new methods to integrate inertial/odometer-based navigation systems with the developed wireless positioning methods for further improvements. The theoretical basis of the work was built on nonlinear stochastic estimation techniques including Particle Filtering, Gaussian Process Regression, Fast Orthogonal Search, Least-Squares, and Radial Basis Functions Neural Networks. All the proposed wireless positioning methods were developed and physically realized on Android-based smart-phones using the IEEE 802.11 WLANs (WiFi). In addition, successful integration with inertial/odometer sensors of mobile robots has been performed on embedded systems. Both theoretical analysis and experimental results showed significant improvements in modeling RSS indoors dynamically without offline training achieving a positioning accuracy of 1-3 meters. Sub-meter accuracy was achieved via integration with inertial/odometer sensors.