A Model for Automatic Extraction of Slowdowns From Traffic Sensor Data

The ability to identify slowdowns from a stream of traffic sensor readings in an automatic fashion is a core building block for any application which incorporates traffic behaviour into its analysis process. The methods proposed in this paper treat slowdowns as valley-shaped data sequences that are found below a normal distribution interval. This paper proposes a model for slowdown identification and partitioning across multiple periods of time and it aims to serve as a first layer of knowledge about the traffic environment. The model can be used to extract the regularities from a set of events of interest with recurring behaviour and to assert the consistency of the extracted patterns. The proposed methods are evaluated using real data collected from highway traffic sensors