Dynamic Prediction of Traffic Congestion by Tracing Feature-Space Trajectory of Sparse Floating Car Data

So-called floating cars (vehicles traveling in traffic, not stationary by the roadside) can be used to collect traffic information without the need for roadside equipment. However, due to the non-stationary distribution of these vehicles, floating-car data (FCD) can be deficient in a number of ways. This becomes a disadvantage when using FCD as a source for prediction, since conventional prediction methods are designed to be used with a continuous data source such as roadside sensor data. This paper reports on the development of a new prediction method that uses tracing feature space trajectory. Feature space is given by principal component analysis on FCD history. Time sequence of deficient FCD describes continuous and cyclic trajectory in the feature space. The continuity and the cyclicity are essential for dynamic prediction, i.e., by tracing the trajectory nearby current coordinates, the method can estimate a prediction point along the trajectory. Inverse projection of the prediction point gives predictive information with no deficiency. The authors report on their evaluation of this method, using FCD collected from 2,000 vehicles to predict traffic congestion. They found that the average error rate of predicted travel time was less than 20%, which is sufficient for practical use.

Language

  • English

Media Info

  • Media Type: CD-ROM
  • Features: Figures; Maps; References;
  • Pagination: 9p
  • Monograph Title: ITS Connections: Saving Time. Saving Lives

Subject/Index Terms

Filing Info

  • Accession Number: 01140789
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Sep 16 2009 4:02PM