Algorithms for map-aided autonomous indoor pedestrian positioning and navigation

Abstract

THESE 3961 (2007) Spassov, Ivan ; ABSTRACT ENG AbstractThe personal positioning and navigation became a verychallenging topic in our dynamic time. The urban canyonsand particularly indoors represent the most difficultareas for personal navigation problematic. Problems likedisturbed satellite signals make the positioningimpossible indoors. Recently developed systems for indoorpositioning do not assure the necessary positioningaccuracy or are very expensive. Our concept stands for afully autonomous positioning and navigation process. Thatis, a method that does not rely on the reception ofexternal information, like satellite or terrestrialsignals. Therefore, this research is based on the use ofinertial measurements of the human walk and the mapdatabase which contains the graphic representation of theelements of the building, created by applying thelink-node model. Using this reduced set of informationthe task is to develop methodology, based on theinteraction of the data from both sources, to assurereliable positioning and navigation process. Thisresearch is divided in three parts. The first partconsists in the development of a methodology for initiallocalization of the person indoors. The problem to solveis to localize the person in the building. Consider aperson equipped with a system which contains set ofinertial sensors and map database of the building. Speed,turn rate and barometric altitude are measured andtime-stamped on each step of the person. A pre-processingphase uses these raw measurements in order to construct apolyline, thus representing user's trajectory. In thelocalization approach central place takes the associationof the user's trajectory with the graph representation ofthe building, process known as map-matching. The solutionis based on statistical method where the determination ofthe user's position is entirely represented by itsprobability density function (PDF) in the frame ofBayesian inference. Initial localization determines theedge of the graph occupied by the person. The second partaims at continuous localization, where user's position isestimated on every step. Besides the application of theclassical map-matching techniques, two new methods aredeveloped. Both rely on the similarity of the geometry ofthe trajectory and the elements of the graph. The firstis based on the Bayesian inference, where the estimationis computed considering the walked distance and azimuth.The second method represents a new application of theFréchet distance as degree of similarity betweentwo polylines. The third part is pointed at thepedestrian guidance. Once the user's position is known itis easy to compute the path to his destination and togive him directions. The problem is to assure continuanceof the process of navigation in the case when the personhas lost his path. In that case the solution consists ineither giving instructions to the user to go back on thepath or computation of a new path from the actualposition of the user to his destination. Based on thatmethodology, algorithms for initial localization,continuous localization, and guidance were created.Numerous tests with the participation of several personshave been provided in order to validate the algorithmsand to show their performance, robustness and limits.