Projet de recherche doctoral numero :4766

Date depot: 1 janvier 1900
Titre: Semantic Technologies for Vehicle Data
Encadrant : Christian BONNET (Eurecom)
Encadrant : Raphael TRONCY (Eurecom)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini

Resumé: Motivation and Research Challenges Sensor data collected from connected vehicles and shared through online services is currently usually limited to speed and location that can be used for traffic estimation. The next step is an extension of sensor data being collected to include pre-defined or recognized speed limits, lanes on the road, slopes, etc. Our vision leads to an autonomous vehicle, where the car will be able to build a detailed environment model based on multiple sensors. The diversity of available sensors between devices, manufacturers, and coming from different ecosystems including personal devices represent a first challenge. While quality of sensors installed in vehicles is increasing, managing different sensor types that have their strengths and weaknesses (“sensor honesty”), that offer different qualities of measurements, and that operate in various contexts represent an important problem. In particular, we will aim to use semantic technologies to provide proper interpretation of the raw data that is being captured. Sensor data can benefit from being integrated with external knowledge bases. However, such a semantic integration is also challenging, in particular when knowledge bases vary constantly like in the case of Open Street Map where crowd-sourced geographic data about new roads and speed limits are constantly updated. A third challenge consists in putting together all sensor data coming from multiple vehicles in a central backend. Fusing data from different vehicles, coming from different manufacturers will increase the need for contextual information. The use of semantic sensor web technologies has the potential to effectively address those challenges. An aspect of growing importance when dealing with data from sensor is data privacy. The vehicle has become a data driven machine. Sharing data between different vehicles and data originating from different drivers require: i) user privacy and awareness, such as transparency and opt-in mechanism, ii) anonymity or pseudonymity of data, as well as technical means to incorporate iii) data provenance and iv) information flow control. Connected Autonomous Vehicles will need to access all kinds of data that are always up-to-date, comprehensive and geographically referenced of its immediate surroundings, as well as cross-OEM crowd-sourced data or real-time data analytics in the cloud. We will investigate methodologies for these data to be communicated and processed between vehicles and the cloud. A relevant research topic in this scenario is data integration at the semantic level while using as much as possible external or crowdsourced knowledge bases. How can such a heterogeneous system with many different stakeholders and completely different sensor types (LIDAR, radar heat, weather,..) work reliably together with data from uncertain sources? How can information about data type, accuracy, freshness and relevance be shared between arbitrary partners within the automotive industry and smart-city data centers? If one sensor is exchanged for a different one, how will the system be able to adapt itself and incorporate a new sensor type? How can knowledge about the interpretation of certain sensor data be shared and exchanged? We propose to use Linked Data technologies combined with the selection and modelling of suitable ontologies as key elements for such a connected system. Linking ontologies from different domains (driving, health, navigation, etc.) will be a key enabler for new classes of use cases making use of the potential of the Internet of Things. Theoretical Background Data privacy and information flow control. A technical necessity for the use of personal and private data in a distributed system is a communication mechanism that knows about data provenance and enforces local and system-wide policies. An approach to establish such communication mechanisms may be borrowed from information security research: information classification and data leakage prevention [1] as well as marshalling of networked data between entities of a distributed sensor networks [2]. Enhancing data with provenance and following its flow through intermediate computation nodes may be an enabler for enhancing privacy in complex and distributed systems. Linked Data and semantic sensor networks The emerging Machine-to-Machine (M2M) field enables machines to communicate with each other without human intervention. Existing semantic sensor networks are domain-specific and add semantics to the context. [3] proposes an architecture that merges heterogeneous sensor networks and add semantics to the measured data rather than to the context. This architecture enables to: (1) get sensor measurements, (2) enrich sensor measurements with semantic web technologies, domain ontologies and the Link Open Data, and (3) reason on these semantic measurements with semantic tools, machine learning algorithms and recommender systems to provide promising applicatio

Doctorant.e: Klotz Benjamin