Daten Lane Change Prediction Using Neural Networks Considering Classwise Non-uniformly Distributed Data
Konferenz: AmE 2018 – Automotive meets Electronics - 9. GMM-Fachtagung
07.03.2018 - 08.03.2018 in Dortmund, Deutschland
Tagungsband: GMM-Fb. 90: AmE 2018
Seiten: 6Sprache: EnglischTyp: PDFPersönliche VDE-Mitglieder erhalten auf diesen Artikel 10% Rabatt
Krueger, Martin; Stockem Novo, Anne; Nattermann, Till; Glander, Karl-Heinz (ZF, Automated Driving & Integral Cognitive Safety, 40547 Düsseldorf, Germany)
Bertram, Torsten (Technische Universität Dortmund, Lehrstuhl für Regelungssystemtechnik, 44227 Dortmund, Germany)
In the operation sequence of automated driving systems, algorithms for situation analysis and interpretation are one important link between environmental perception and trajectory planning. The more accurately known the current traffic situation is and the more precisely estimated the possible future evolution of this traffic situation can be, using situation interpretation methods, the more foresighted the vehicle handling will be. Since the strict structuring of road and lane configurations in highway scenarios allows only a limited variety of driving maneuvers, the information whether a neighboring vehicle is currently performing a lane change or if a lane change is impending is already valuable. Many different machine learning methods like neural networks are prone to imbalanced datasets. Datasets recorded for lane change prediction are usually imbalanced. In this contribution, this aspect of the training of neural networks during supervised learning is investigated.