In recent years, the use of emergency management systems has become increasingly widespread. Recently, the European Union started the Evolution of Emergency Copernicus Services project to enhance the capabilities of its emergency management system, the Copernicus Emergency Management Service. This thesis focuses on the enhancement of its Rapid Mapping component, that can produce detailed maps of an area involved by an emergency event. Since it was common to experience delays on maps production due to satellite availability issues, it was introduced the Copernicus Witness Service Component. This tool is able to retrieve additional geographical data to use for mapping purposes by geolocating the text of social media posts related to the considered emergency event. This thesis developed a methodology to improve the Copernicus Witness Service Component by making it capable to enhance the geolocation of each collected social media post by analyzing also their media, if available. A set of tools based on deep learning and computer vision techniques were developed. These tools are able to recognize relevant images, to assess their point of view and to compare them with georeferenced imagery of the area involved by the emergency event to improve their geolocation. Finally, it was developed a classification score based on all the collected information to present to emergency mappers relevant social media posts by prioritizing the one that are geolocated more precisely.
Negli ultimi anni, l'uso degli emergency management systems è diventato sempre più diffuso. Recentemente, l'Unione Europea ha avviato il progetto Evolution of Emergency Copernicus Services per migliorare i servizi del suo emergency management systems, il Copernicus Emergency Management Service. Questa tesi nello specifico si occupa di migliorare le prestazioni del suo componente Rapid Mapping, in grado di produrre mappe dettagliate di un'area coinvolta da un disastro naturale. Poiché era comune riscontrare ritardi nella produzione delle mappe a causa di problemi di disponibilità dei satelliti, è stato introdotto il Copernicus Witness Service Component. Questo strumento è in grado di recuperare ulteriori dati geografici da utilizzare per la produzione di mappe di emergenza geolocalizzando il testo dei messaggi dei social media relativi al disastro considerato. Questa tesi propone una metodologia che va a migliorare il Copernicus Witness Service Component rendendolo in grado di affinare la geolocalizzazione di ogni post raccolto sui social media analizzando anche i suoi media, se disponibili. Per fare ciò è stato sviluppato un insieme di strumenti basati sul deep learning e su tecniche di computer vision. Questi strumenti sono in grado di riconoscere le immagini rilevanti, di valutarne il punto di vista e di confrontarle con le immagini georeferenziate dell'area interessata dall'evento di emergenza per migliorarne la geolocalizzazione. Infine, è stato sviluppato un metodo di classificazione dei post basato su tutte le informazioni raccolte da questi strumenti, in modo da presentare a coloro che producono le mappe di emergenza i post dei social media dando la priorità a quelli più rilevanti, ossia a quelli geolocalizzati in modo più preciso.
A methodology for the geolocation of social media information in the context of emergency mapping
BALDASSARI, ALESSANDRO
2018/2019
Abstract
In recent years, the use of emergency management systems has become increasingly widespread. Recently, the European Union started the Evolution of Emergency Copernicus Services project to enhance the capabilities of its emergency management system, the Copernicus Emergency Management Service. This thesis focuses on the enhancement of its Rapid Mapping component, that can produce detailed maps of an area involved by an emergency event. Since it was common to experience delays on maps production due to satellite availability issues, it was introduced the Copernicus Witness Service Component. This tool is able to retrieve additional geographical data to use for mapping purposes by geolocating the text of social media posts related to the considered emergency event. This thesis developed a methodology to improve the Copernicus Witness Service Component by making it capable to enhance the geolocation of each collected social media post by analyzing also their media, if available. A set of tools based on deep learning and computer vision techniques were developed. These tools are able to recognize relevant images, to assess their point of view and to compare them with georeferenced imagery of the area involved by the emergency event to improve their geolocation. Finally, it was developed a classification score based on all the collected information to present to emergency mappers relevant social media posts by prioritizing the one that are geolocated more precisely.| File | Dimensione | Formato | |
|---|---|---|---|
|
A methodology for the geolocation of social media information in the context of emergency mapping - Alessandro Baldassari - 841561.pdf
non accessibile
Descrizione: Testo della tesi - Revisione dopo consegna 8 luglio 2019
Dimensione
6.56 MB
Formato
Adobe PDF
|
6.56 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/148567