Thibaud Michel - On Mobile Augmented Reality Applications based on Geolocation

Organized by: 
Thibaud Michel
Thibaud Michel


This thesis was written as part of a collaboration between two teams: Tyrex (LIG / INRIA) and Necs (Gipsa / INRIA).

Jury :

  • Valérie Renaudin, chargé de recherche, IFFSTAR; Reviewer
  • Takeshi Kurata, Professor, AIST; Reviewer
  • Laurence Nigay, professeur; Examiner
  • Pierre Genevès, chargé de recherche, CNRS; Supervisor
  • Hassen Fourati, maître de conférences, Gipsa-lab; Supervisor
  • Nabil Layaïda, directeur de recherche, Inria; Supervisor

Applications for augmented reality can be designed in various ways, but few take advantage of geolocation. However, nowadays, with the many cheap sensors embedded in smartphones and tablets, using geolocation for augmented reality (Geo AR) seems to be very promising. In this work, we have contributed on several aspects of Geo AR: estimation of device positioning and attitude, and the impact of these estimations on the rendering of virtual information.

In a first step, we focused on smartphone attitude estimation. We proposed the first benchmark using a motion lab with a high precision for the purpose of comparing and evaluating filters from the literature on a common basis. This allowed us to provide the first in-depth comparative analysis in this context. In particular, we focused on typical motions of smartphones when carried by pedestrians. Furthermore, we proposed a new technique for limiting the impact of magnetic perturbations with any attitude estimation algorithm used in this context. We showed how our technique compares and improves over previous works.

In a second step, we studied the estimation of the smartphone's position when the device is held by a pedestrian. Although many earlier works focused on the evaluation of localisation systems, it remains very difficult to find a benchmark to compare technologies in the setting of a commodity smartphone. Once again, we proposed a novel benchmark to analyse localisation technologies including WiFi fingerprinting, WiFi trilateration, SHS (Step and Heading System) and map-matching.

In a third step, we proposed a method for characterizing the impact of attitude and position estimations on the rendering of virtual features. This made it possible to identify criteria to better understand the limits of Geo AR for different use cases.

We finally proposed a framework to facilitate the design of Geo AR applications. We show how geodata can be used for AR applications. We proposed a new semantics that extends the data structures of OpenStreetMap. We built a viewer to display virtual elements over the camera livestream. The framework integrates modules for geolocation, attitude estimation, POIs management, geofencing, spatialized audio, 2.5D rendering and AR. Three Geo AR applications have been implemented using this framework. TyrAr is an application to display information on mountain summits and cities around the user. AmiAr allows one to monitor lights, shutters, tv in a smart apartment. Venturi Y3 is an AR-Tour of Grenoble with audio description and experiences.