Tales Paiva Nogueira - A Framework for Automatic Annotation of Semantic Trajectories

12:30
Monday
30
Jan
2017
Place: 
Organized by: 
Tales Paiva Nogueira
Speaker: 
Tales Paiva Nogueira
Teams: 
Detailed information: 

Composition du jury :

  • M Thomas Devogele, professeur, Université de Tours, rapporteur
  • M Alain Bouju, maître de conférences HDR, Université de La Rochelle, rapporteur
  • M Hervé Martin, professeur, Université Grenoble Alpes, directeur de thèse
  • M Ahmed Lbath, professeur, Université Grenoble Alpes, examinateur
  • Mme Rossana Maria de Castro Andrade, professeur, Federal University of Ceará, Brésil, examinateur
Abstract: 

Location data is ubiquitous in many aspects of our lives. We are witnessing an increasing usage of this kind of data by a variety of applications. As a consequence, information systems are required to deal with large datasets containing raw data in order to build high level abstractions. Semantic Web technologies offers powerful representation tools for pervasive applications. The convergence of location-based services and Semantic Web standards allows an easier interlinking and annotation of trajectories.
    
In this thesis, we focus in modeling mobile object trajectories in the context of the Semantic Web. First, we propose an ontology that allows the representation of generic episodes. Our model also handles contextual elements that may be related to trajectories. Second, we propose a framework containing three algorithms for automatic annotation of trajectories. The first one detects moves, stops, and noisy data; the second one is able to compress generic time series and create episodes that resumes the evolution of trajectory characteristics; the third one exploits the linked data cloud to annotate trajectories with geographic elements that intersects it with data from OpenStreetMap.
    
As results of this thesis, we have a new ontology that can represent spatiotemporal phenomena at different levels of granularity. Moreover, our framework offers three novel algorithms for trajectory annotation. The move-stop-noise detection method is able to deal with irregularly sampled traces and do not depend on external data of the underlying geography; our time series compression method is able to find values that summarize a series at the same time that too small segments are avoided; and our spatial annotation algorithm explores linked data and the relationships among concepts to find relevant types of spatial features to describe the environment where the trajectory took place.