Classifying Road Intersections Using Transfer-Learning on a Deep Neural Network

Ulrich Baumann, Yuan-Yao Huang, Claudius Gläser, Michael Herman, Holger Banzhaf, J. Marius Zöllner

IEEE 21th International Conference on Intelligent Transportation Systems, 2018, Maui, Hawaii, USA, November 4 – 7, 2018

Abstract:

With the steady rise of advanced driver assistance systems (ADAS), more and more aspects of the driving task are transferred from the human driver to the vehicle’s control system. In order to handle many of these responsibilities, vehicles need to understand their environment and adjust their behavior according to it. An important aspect of the vehicle environment is the layout of the road segment right ahead of the vehicle, such as the presence and type of an intersection, as it defines the scenario, provides context information and constrains the future motion of traffic participants. The knowledge of upcoming intersections can help to improve various aspects in the context of driver assistance systems and automated driving, such as the prediction of traffic participants or the adjustment of a system with respect to the current scenario. The contribution of this paper is threefold: First, it introduces a model for intersection identification and classification ahead of a vehicle solely from on-board sensor data via deep learning. Second, it proposes a transfer-learning technique allowing to train with fewer samples and showing that intermediate features from path prediction are also beneficial for intersection classification tasks. Third, it allows to reduce necessary computational power since feature extraction is partially shared between the path prediction and the intersection classification model.

PDF Bibtex

@INPROCEEDINGS{Baumann2018ITSC,
  author={U. {Baumann} and Y. {Huang} and C. {Gläser} and M. {Herman} and H. {Banzhaf} and J. M. {Zöllner}},
  booktitle={2018 21st International Conference on Intelligent Transportation Systems (ITSC)},
  title={Classifying Road Intersections Using Transfer-Learning on a Deep Neural Network},
  year={2018},
  month={November},
  pages={683-690}, 
}