Dissertation

Representativeness Analysis of Driving Data

Author:

Johannes Hiller

Pages:

218

DOI:

10.18154/RWTH-2024-06388

Year:

2024

Language:

english

Format:

ebook

With the further development of automated vehicles, safety assurance as a step towards the introduction on public roads gains in importance. Typically, the approach involves the analysis of recorded driving data in specific situations, so-called scenarios. Scenarios concerning the interaction between traffic participants are often considered the most relevant. However, with the further maturation of the vehicles, other traffic influencing factors come into focus including environmental factors such as road or weather conditions. For both interaction and environmental factors, the question of balancing the influences remains. This thesis aims at answering this question with the calculation of the representativeness of parameter values regarding one or multiple references.

In order to achieve this, a framework is designed that incorporates three major stages: aggregation of reference data to parameter spaces, enrichment of recorded driving data and finally the calculation of the representativeness based on the two previous stages.

As reference data, map data from OpenStreetMap and weather data from Deutscher Wetterdienst are used. They are aggregated based on distance (only for map data) and by matching them to traffic volume and accident data. For the second stage, the driving data are provided in a specified common data format, but the environmental data are optional as they are not always needed in today’s safety assurance applications. Therefore, the enrichment of the driving data plays an important role in the process for the calculation of the representativeness. Using the same data sources as for the aggregation, the enrichment is done based on map-matching algorithms for the map data and via a lookup scheme for the weather data. Combined with the subsequent driving scenario detection, the parameter spaces for the calculation of the representativeness are available.

The calculation of the representativeness is the final stage of the proposed framework and uses the two previous steps as input. A method is applied which allows setting parameters (e.g., speed limit of 100 km/h) or combinations of parameters (e.g., speed limit of 100 km/h on two-lane motorway and rain) into relation. Based on these relations, the representativeness is calculated. Utilizing the results, it is possible to state the overor underrepresentation of parameter values in the driving data. Formulating an optimization problem, this allows the calculation of the additionally required kilometers to be recorded in order to balance the dataset. Using the presented modular method, an extension to further data sources and reference is perceivable.