Methods for the continuous network-wide quality analysis of road traffic flow

点击翻译按钮获取中文摘要

The network-wide analysis of the level of service (LOS) of car traffic within a road network should be one of the regular tasks of traffic management in order to either maintain a stable traffic flow or to identify and cure bottlenecks. With respect to nowadays strictly limited resources, road authorities in charge are not able to pursue this important goals, as traditional sensor techniques (i.e. loop detectors) and analytical LOS equations only allow a very limited view on the road network. During the recent years the increasing availability of floating car data (FCD) caused by the popular use of smartphone navigation software and modern on-board-navigation systems offer a high potential to overcome the actual situation. The data analysis of anonymized FCD trajectories enables traffic planners to measure directly and highly cost efficient important LOS indicators, like delay times, tailback lengths and link or route travel times. During the last decades several research regarding different FCD-based LOS detection techniques has been proposed in literature and were tested in reality. The most important and critical aspect of these methods could be seen in the data provisioning and model validation, as most methodologies require a very dense sampling resolution and high FCD penetration rates, to be able to calculate representative LOS estimates. Only the minority of research fits to typical low-frequency data, which could be supplied by FCD providers nowadays. FCD penetration rate varies highly between intra and extra urban areas, which implies different data analysis methodologies, that allow an intelligent adaptation to available data provisioning. The main target of this research proposal is therefore, the development of validated LOS estimation techniques, based on realistic low-frequency FCD provided by GARMIN/Navigon. Furthermore, the complete processing of very large FCD archives requires a very complex stack of data processing methodologies, that cover aspects of graph theory, data filtering, path inference and map-matching, whereby the applied processing approaches by itself influence the data output and therefore the representativity of estimated LOS indicators. To be able to simplify the work on FCD and to guarantee comparable indicators, this research project additionally aims on the development of a validated FCD processing stack, to give further research on FCD analysis an easier entry point.