Abstract:
This dissertation presents a novel process for generating lane-level information for road maps using a collection of trajectories from vehicles travelling on the road. The information thus generated will aid a range of Intelligent Transportation System (ITS) applications. Recently, there has been a surge of interest in research in the arena of Intelligent Transportation Systems (ITS). These systems are expected to ensure driver/passenger safety, assist the driver, support green concepts and to improve the overall efficiency and the performance of transportation systems. However, ITS require more information than what current road maps provide. Generating and refining road maps having such level of detail using the existing methods such as surveying and digitization are time consuming, costly and incompatible with the real-time and the dynamic nature of the road network. Therefore, finding new ways of generating this additional information is of high importance in making ITS a reality. The new method we propose generates lane-level information such as lane centerlines, boundaries and lane width using vehicle trajectory data. This is achieved by modeling the Probability Density Function (PDF) of trajectories across the road using the non-parametric Kernel Density Estimation (KDE). Unlike the existing methods that use Differential GPS (DGPS) data or improved GPS data, the proposed method uses ordinary GPS data obtained from vehicles moving along the road. It does not require any information regarding the road parameters and is completely automatic. Furthermore, it is completely independent of the lane/road width and does not use stringent assumptions on lane parallelism and constant lane width. In particular, it estimates the locations of lane centers, locations of lane boundaries and lane width. The proposed method for calculating the lane centers was proven to be successful in different road geometries such as straight sections, curved sections and sections with lane splits and merges. The method proposed for calculating lane boundaries produced good results when there are no gaps in between lanes. The lane width calculated using the proposed method is compatible with the recorded standard lane with of the chosen road. Keywords: GPS, Lane level maps, Kernel density estimation, Kernel bandwidth
Citation:
Manel, E.R.I.A.C. (2013). Probabilistic estimation of lanes using vehicle GPS trajectories [Master's theses, University of Moratuwa]. Institutional Repository University of Moratuwa. http://dl.lib.mrt.ac.lk/handle/123/10685