M.Sc Thesis

M.Sc StudentBenaim Shay
SubjectHierarchical Approach for Extracting Road Network from
Crowdsourced Data
DepartmentDepartment of Civil and Environmental Engineering
Supervisor DR. Sagi Dalyot
Full Thesis text - in Hebrew Full thesis text - Hebrew Version


In recent years, technology related to information sharing has evolved. One of its unique outcomes is the option of sharing geographic information, mostly by the internet, where users share data about their position, and other additional information. One type of geographic information, which can be contributed, is GPS trajectories of road users (pedestrians, cyclist, car drives, etc.). GPS trajectories are collected while driving a vehicle, whereas the cellular phone is sampling the way the user is traveling, by creating a series of locations with time stamps.

This research presents the potential of using road GPS trajectories for the automatic creation of vector road map. Two main issues have to be solved due to the use of road trajectory data, which represent roads:

1.      Linking between several road trajectories that were recorded on the same road: several road trajectories, which were recorded on the same road, most evidently will have different global positioning, due to GPS receiver inaccuracies, mostly related to different quality, satellite constellation and road conditions.

2.      Integration of several road trajectories that represent the same road segment to a probable vector road: since the road segment has several road trajectories not having knowledge on the traces’ accuracy, road condition and the quality of the GPS receiver, there exists a difficulty to determine the probable vector road.

To resolve the research goal, a hierarchic methodology was developed, composed of these main processes:

1.    Extracting the junction nodes from the road trajectories.

2.    Divide the road trajectories to road segments, which are defined as the way between tow junction nodes.

3.    Integrate the trajectories segments to a single and unified road vector.

One conclusion of this research is that for different road characterizations, such as width (number of lanes), usage, traffic volume and trajectories volume (data analysed), different sets of parameters used in the algorithms are required. Moreover, the examination of the road creation process shows that it is suitable for all kind of road geometry, but it is sensitive for the number of junctions it is described by. However, the hierarchic process demonstrates a fully automatic process, enabling the extraction of road directions, mapping of road geometry and stopping points (mostly junctions).

This new approach for automatic road mapping that considers the road characteristics enables accurate mapping of road network with no need of a-priori information. The suggested hierarchic process shortens and optimizes the commonly implemented road mapping process (surveying, digitization), while improving the mapping quality, making use of geospatial information, which is available online and for free, retrieved from accurate and qualitative commonly used sensors.