A probabilistic framework for traffic data quality

Abstract

Regarding the assessment of traffic data in ITS, there is an increasing need for answers to the following questions: (i) What exactly is "traffic data quality"?, and, related to that, (ii) There are too many ways to define and to do things, and results of different researchers are inconsistent or not comparable. How can we overcome this situation? With that background, an important aim of the ongoing DLR-project I.MoVe is to develop a consistent understanding of traffic data quality, together with a unified framework for its assessment. To this end, a probabilistic framework for traffic data quality is provided in this paper. Real-world examples from I.MoVe demonstrate its application for the assessment of data sources like induction loops, stationary bluetooth sensors and floating car data (FCD). A first important point is to distinguish strictly between quality indices, quality requirements, and quality itself. While the present framework develops quality indices based on established quality criteria like accuracy, completeness, validity, and coverage, the usual understanding of quality is extended to a probabilistic view. This also addresses the problem of information retrieval in the presence of vagueness and uncertainty. The provided examples are making full use of the proposed framework, and also constitute interesting results for the practitioner by themselves. Examples include the assessment of induction loop count data, and assessing the temporal coverage of a stretch of road with stationary bluetooth data, or of the whole city of Berlin, Germany with FCD.