Data Stream Processing Method for Clustering of Trajectories

Abstract

The constant advances in techniques for recording and collecting GPS trajectory information, the increase in the number of devices that collect this type of information such as video cameras, traffic sensors, smart phones, etc., has resulted in a large volume of information. Being able to process this information through data streams that allow intelligent analysis of the data in real time is an area where many researchers are currently making efforts to identify solutions. GPS trajectory clustering techniques allow the identification of vehicle patterns over large volumes of data. This paper presents a method that processes data streams for dynamic clustering of vehicular GPS trajectories. The proposed method here receives a GPS data stream, processes it using a buffer memory and the creation of a grid with the use of indexes, and subsequently analyzes each cell of the grid with the use of a dynamic clustering technique that extracts the characteristics of reduced zones of the study area, visualizing common speed ranges in interactive maps. To validate the proposed method, two data sets from Rome-Italy and Guayaquil-Ecuador were used, and measurements were made of execution time, used memory and silhouette coefficient. The obtained results are satisfactory.