Name
Abstract | ||
|---|---|---|
Predicting driver behavior in general, and the problem of predicting an impending lane change in particular have been studied in the community under different aspects and setups. Mostly the problem is approached as a classification problem in which each class represents a possible action in the next few seconds. In this work, we re-define the task as a regression problem, in which the time until the ego-vehicle touches the other lane is predicted. In particular, we use Long-Short-Term-Memory (LSTM) networks to learn to predict the time to lane change. Even though a regression-based formulation of the problem captures more information and should thus be harder to learn, we can show that it provides slightly better results than a comparable classification-based approach. Moreover, the additional precision that we get about the exact time of the impending the lane change could be used to further increase a user's acceptance of advanced driver assistance systems. We also show that it is possible to individualize the network by fine-tuning it to a particular driver's behavior. The fine-tuning results in an improvement of F1-Score while observing about 20 minutes of driving data from that driver. |
Year | DOI | Venue |
|---|---|---|
2017 | 10.1109/ITSC.2017.8317674 | 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC) |
Keywords | Field | DocType |
time-to-lane-change prediction,deep learning,predicting driver behavior,impending lane change,classification problem,regression problem,Long-Short-Term-Memory networks,advanced driver assistance systems,ego-vehicle,regression-based formulation,user acceptance,F1-Score,fine-tuning | Regression,Simulation,Advanced driver assistance systems,Artificial intelligence,Deep learning,Engineering,Regression problems,Change prediction,Machine learning | Conference |
ISSN | ISBN | Citations |
2153-0009 | 978-1-5386-1527-0 | 3 |
PageRank | References | Authors |
0.46 | 5 | 4 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Hien Q. Dang | 1 | 3 | 1.14 |
| Johannes Fürnkranz | 2 | 2476 | 222.90 |
| Alexander Biedermann | 3 | 3 | 0.46 |
| Maximilian Höpfl | 4 | 3 | 0.46 |