Name
Abstract | ||
|---|---|---|
In this paper, we present a hybrid classifier built by combining object detection (YOLOv5) and signal aspect detection through traditional image processing routines. The hybrid classifier is part of our test automation toolchain. It is intended to be used for monitoring railway signals in end-to-end integration test scenarios when bringing interlockings into service. In our field tests, we have successfully demonstrated recognition and classification of signal aspects for German KS signals and UK running signals. The aspect detection provides distance metrics for the position, radius, and color of the lights in the signal. We trained a logistic regression classifier with these metrics to classify detected signal light candidates. The set of detected lights is then matched against all predefined signal aspects to determine the signal aspect. We tested our algorithm in our lab and a test field. The F1 score of the presented algorithm is 0.74 for the KS signal and 1 for the UK running signal. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1109/EAIS51927.2022.9787732 | 2022 IEEE International Conference on Evolving and Adaptive Intelligent Systems (EAIS) |
Keywords | DocType | ISSN |
signal aspect detection hybrid classifier computer vision object detection railway infrastructure railway digitalization | Conference | 2330-4863 |
ISBN | Citations | PageRank |
978-1-6654-3707-3 | 0 | 0.34 |
References | Authors | |
6 | 3 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Dirk Friedenberger | 1 | 0 | 0.34 |
| Felix Grzelka | 2 | 0 | 0.34 |
| Andreas Polze | 3 | 268 | 51.57 |