Authors:
Carl Esselborn
1
;
Leo Misera
1
;
Michael Eckert
1
;
Marc Holzäpfel
1
and
Eric Sax
2
Affiliations:
1
Dr. Ing. h.c. F. Porsche AG, Weissach, Germany
;
2
Department of Electrical Engineering, Karlsruhe Institute of Technology, Karlsruhe, Germany
Keyword(s):
Map Data, Validation, Floating Car Data, Anomaly Detection, Autoencoder, Yield Signs.
Abstract:
Map data is commonly used as input for Advanced Driver Assistance Systems (ADAS) and Automated Driving (AD) functions. While most hardware and software components are not changed after releasing the system to the customer, map data are often updated on a regular basis. Since the map information can have a significant influence on the function’s behavior, we identified the need to be able to evaluate the function’s performance with updated map data. In this work, we propose a novel approach for map data regression tests in order to evaluate specific map features using a database of historic floating car data (FCD) as a reference. We use anomaly detection methods to identify situations in which floating car data and map data do not fit together. As proof of concept, we applied this approach to a specific use case finding yield signs in the map, which are currently not present in the real world. For this anomaly detection task, the autoencoder shows a high precision of 90% while maintai
ning an estimated recall of 45%.
(More)