Big data computing for traffic information by GPS sensing

Real time traffic monitoring systems perform spatial and time dependent analysis of measurement data of different types such as traditional inductive loop detector data, microwave radar data, and GPS data. The goal of these systems is to provide information such as average speeds, volumes and densities on a given segment of a roadway. One of the fastest growing data source for traffic monitoring systems is GPS data collected from mobile devices. To some extent, in the industry GPS data is already replacing the traditional traffic sensing technologies. There is a large demand in industry and transportation agencies to have access to high resolution state of traffic on highways and arterial roads globally. This means that traffic information providers have to provide traffic information on a resolution that goes beyond the widely used TMC code based representation of the roadway. In order to obtain the high resolution state of traffic, noisy observations need to be fused into a mathematical model that represents the evolution of the system either based on physics or statistics. Common frameworks for fusing the data into physical models include for example Kalman filtering and particle filtering. Prior to the data fusion stage in the real time system, offline geospatial modelling has already been done. For example, construction and calibration of an accurate physics based traffic model includes tasks such as building a directed graph of the road network, detection of road geometry at lane level and speed limit detection. In all these tasks, GPS data is vital. Real time systems that use GPS data include geospatial pre-processing components such as map matching and path inference. The rapidly growing volume of GPS data cannot be handled using traditional methods but instead parallel computing systems are needed to handle the future volumes. Also, the new high resolution algorithms are developed to leverage the parallel processing frameworks. In this talk I will discuss directions taken to respond to the demand of providing high resolution information about the state of the traffic both in the context of modeling and implementation of a large scale system.