Information Enrichment of POIs using GPS trace data

From Master Projects
Jump to: navigation, search


has title::Automatic Parking Area Detection using GPS trace data
status: finished
Master: project within::Technical Artificial Intelligence
Student name: student name::Victor Anchidin
Dates
Start start date:=2011/02/01
End end date:=2011/07/11
Supervision
Supervisor: Zoltán Szlávik
Second reader: has second reader::Wojtek Kowalczyk
Company: has company::TomTom
Poster: [[has poster::Media:[[1]]]]

Signature supervisor



..................................

Abstract

TomTom is the world’s leading provider of location and navigation solutions. The dedicated portable navigation devices are used to collect a real-time feedback from users in order to find valuable information about the traffic conditions and locations, which will be spread later to users as a dynamic content.

Everyday a lot of information about Points Of Interests (POIs) is accessed by millions of people from all over the world. Knowledge like popularity, opening hours and accessibility can improve the user experience, but it is also valuable for the owner of the place.

The GPS trace data acquired from millions of connected navigation devices contains a wealth information about when and how users navigate to their destination. However the currently available GPS traces also have their limitations. Most of the data comes from car navigation, while most places do not have a dedicated parking place. Therefore, the challenging part would be to map traces to an individual POI. In addition, the GPS traces are completely anonymized, so it is impossible to relate traces from different days to the same user or vehicle.

The current project will study the possibilities to use GPS trace data for knowledge enrichment of the POI database. It will present a prototype system that automatically clusters GPS data that was acquired during a long period of time from an extended region. The clustering methods will split the traces according to their destinations seen at multiple scales. The results of the clustering methods will be presented as heat maps and will be extensively analyzed.

Research needs to be performed to study the current state-of the art work related to POIs and GPS traces. An overview needs to be made to indicate possible applications of GPS data for POI enrichment and which types of POIs can benefit from trace information. Further analysis should be done in order to find out the if the results can be extended to other countries and the amounts of data necessary.

KIM Abstract

TomTom is the world’s leading provider of location and navigation solutions. The dedicated portable navigation devices are used to collect a real-time feedback from users in order to find valuable information about the traffic conditions and locations, which will be spread later to users as a dynamic content. During the KIM presentation, some limited insights about the TomTom technologies will be presented, and how they map to the current project.

Everyday a lot of information about Points Of Interests (POIs) is accessed by millions of people from all over the world. Knowledge like popularity, opening hours and accessibility can improve the user experience, but it is also valuable for the owner of the place. During the KIM presentation, the goal of the project will be linked to its content and to the technologies used in TomTom.

The GPS trace data acquired from millions of connected navigation devices contains a wealth information about when and how users navigate to their destination. However the currently available GPS traces also have their limitations. Most of the data comes from car navigation, while most places do not have a dedicated parking place. Therefore, the project is expected to map the traces to individual parking places and to extract some valuable information about these parking places. During the KIM presentation, the identified issues and some potential solutions will be presented. Moreover, some limitations of the GPS traces and current data will be discussed.

The project studies the possibilities to use GPS trace data for knowledge enrichment of the POI database, with particular attention to parking places. It will present a prototype system that automatically clusters GPS data that was acquired during a long period of time from an extended region. The clustering method will split the traces according to their destinations seen at multiple scales. During the KIM presentation, the methods will be discussed and some validation procedures that will be implemented will be presented. In the end, the audience will have a clear overview of how the project will be evaluated.

Research has to be performed to study the current state-of the art work related to POIs and GPS traces. An overview needs to be made to indicate possible applications of GPS data for POI enrichment and which types of POIs can benefit from trace information. Further analysis should be done in order to find out whether the results can be extended to other countries, and the amounts of data necessary to do so. During the KIM presentation, there will be discussed the expected results and a short plan for the next period in implementing the prototype and writing the master thesis.

Final KIM Abstract

Navigation devices have the ability to act as intelligent agents that assist users in various spatial contexts defined by their geographic position. Previous work has explored the possibilities of important place extraction from raw GPS data. However, they all have limitations imposed especially by the data they use. This thesis proposes a framework that is able to automatically discover meaningful parking places directly from raw GPS logs acquired from users of TomTom navigation devices. The framework contains fast and robust algorithms for staying point extraction and clustering methods for parking place identification. The developed clustering methods are tuned with various ground truth datasets. The proposed framework is validated using the data consisting of one month of GPS logs acquired from the TomTom drivers in the West of the Netherlands. The proposed method proves to be able to extract a large number of previously unknown parking areas from the data.