With the vast amount of potential traffic scenarios, the identification of certain patterns is key to ensure a broad scope, while minimizing the effort in the validation process for autonomous driving functions. An expert driven search for such patterns is laborious and likely to be incomplete. Car manufacturers and other parties collect data continuously, so that machine learning models are promising approaches to support engineers and researchers. Such models should be able to recognize patterns without supervision, since supervision requires expert-guided labels, which brings one back to the initial problem. Hence, the scope of this chapter is to introduce an unsupervised learning method for the categorization of traffic scenarios. The method is based on Random Forests and performs the pattern recognition only given the input from arbitrary data sources.
Unsupervised Random Forest Learning for Traffic Scenario Categorization
Machine Learning and Optimization Techniques for Automotive Cyber-Physical Systems ; Kapitel : 20 ; 565-590
2023-09-02
26 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Unsupervised image categorization
British Library Online Contents | 2005
|Ensemble Partitioning for Unsupervised Image Categorization
British Library Conference Proceedings | 2012
|Springer Verlag | 2011
|