Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Tagging of incidents from scattered historical large-scale traffic data sets

: Förster, G.; Klingner, M.; Krimmling, J.

Martin, U. ; Univ. Stuttgart:
Networks for Mobility 2008. Abstracts and CD-ROM : Proceedings of the 4th International Symposium, September 25-26, 2008, Stuttgart, Germany
Stuttgart, 2008
ISBN: 978-3-921882-24-5
9 S.
International Symposium "Networks for Mobility" <4, 2008, Stuttgart>
Fraunhofer IVI ()
incident detection; multivariate data analysis; principal component analysis; Singular Value Decomposition; traffic state estimation; floating car data; inductive loop detector; traffic monitioring system

Principal component analysis (PCA) provides a smart and very efficient technique to extract incident signals from noisy traffic data. Even though the idea of PCA is not very hard to understand, it is relatively uncommon to apply it for traffic data analysis yet. PCA does not necessarily focus either on the temporal or on the spatial dimension of traffic data only. But rather, the whole data space can be rotated simultaneously and two-dimensional weights can be produced. By interpreting these weights specific anomalies in the data set can be identified. In the present paper, first the characteristics of inductive ILD and FCD measurements in generally and with respect to the test site are discussed. After a short methodological introduction into PCA some references for its application are presented. Then the four-month period test data sets of both different sources are evaluated by PCA. It is shown how particular incidents effect on the PCA-weights and how these weights can be interpreted. The paper ends with future prospects on options for online-application of PCA for traffic incident detection.