ABSTRACT
It is crucial for navigation systems and all other map-based applications to obtain highly accurate and quickly updated road maps. In recent years, with the popularization of GPS equipment on mobile objects such as vehicles and mobile phones, the spatiotemporal trajectories of such objects are becoming available. This chapter aims at proposing an algorithm for recognizing high-accuracy road centerlines by using large-scale but coarse-grained GPS trace data. Emphasis is on improving the accuracy of road centerlines. The chapter introduces the preliminaries of using original data to filter out abnormal data and partition the original traces. It then discusses the algorithm for generating maps in detail and presents the effectiveness of applying LS-SVM to fit the centerlines. The chapter also introduces the process of data cleaning to filter out corrupt data. Then a dedicated partitioning process is created to filter out some useless data and to reduce the input to the road-recognition algorithm.
