Mining road map from big database of GPS data

Abstract

This paper describes a process of converting raw Global Positioning System (GPS) data to a routable road map. In fact, it is a large scale database collected from thousands of vehicles circulating on Tunisian public roads.Moreover, the paper contains the architecture used to collect GPS data from these vehicles using GPRS connection and all the steps until getting the road traces.The data flow is composed of many steps which are: Collecting data which consists of extracting National Marine Electronics Association(NMEA) sentences; Filtering raw GPS nodes to eliminate outliers and noise caused by several sources of errors; Clustering step, in which we used two methods partitional (k-means)and hierarchical (agglomerative)clustering techniques. We compare them and we choose the most suitable for our work. In fact, K-means algorithm is carried out in order to partition data and facilitate handling the big data sets; Generating a Tunisian map network from our database and map-matching it with Google maps in order to make a comparison between them.

Collection and exploration of GPS based vehicle traces database

“…In fact, data were gathered from more than 10000 GPS receivers mounted in many fleet of vehicles. Different kinds of vehicles are used in the collection step [8]. More over, they were moving along the Tunisian domain in four years (2009)(2010)(2011)(2012)(2013)(2014).…”

“…The classification method is K-means thanks to its benefits [8]. In our case, we can define the number of clusters as we need according to cluster size and the capacity of hardware available to process GPS data.…”