Annotation of Traffic Scenarios for Autonomous Drive Verification Using Active Learning
- 提供方法
- 本サイト上にてダウンロード・閲覧可
- 形態
- 価格
- 一般価格(税込):¥1,100 会員価格(税込):¥880
- 文献番号
- 20219029
- 文献・情報種別
- その他の国際会議
- 掲載ページ
- 1-6(Total 6 p)
- 発行年月
- 2021年 9月
- 出版社
- (公社)自動車技術会
- 言語
- 英語
書誌事項
著者(英) | 1) Sanna Jarl, 2) Julia Wennerblom, 3) Maria Svedlund, 4) Sadegh Rahrovani, 5) Morteza Hagir Chehreghani |
---|---|
勤務先(英) | 1) Chalmers University of Technology, 2) Chalmers University of Technology, 3) Volvo Cars Coorporation, 4) Volvo Cars Coorporation, 5) Chalmers University of Technology |
抄録(英) | New business models and digitalization have revolutionized the automotive sector, as many other industries. Self-driving vehicles are considered to be one mega trend that is of high importance for the future of automotive industry. In order to comfortably integrate Autonomous Drive (AD) in society, it must be safe and tested properly; a demand that requires extracting and annotating large amounts of driving scenarios for verification and validation of autonomous cars. However, annotating the driving scenarios based only on explicit rules (i.e., knowledge-based methods) can be subject to errors, such as false positive/negative classification of scenarios that lie on the border of two scenario classes, missing unknown scenario classes and anomalies. On the other side, verifying labels by the annotators is not cost-efficient. For this purpose, active learning is a paradigm which increases annotation accuracy by efficient inclusion of an annotator/expert. A classifier is trained on a small labelled dataset, and the bulk of unlabeled data is then classified. The label of the most informative data samples is queried, for instance by an expert, added to the annotated dataset whereafter the classifier is retrained. The goal of this study is to test the performance of active learning on time series data, coming form vehicle motion trajectories. To model the temporal nature of data, we consider three latent space representations: multivariate Time Series t-Distributed Stochastic Neighbor Embedding (mTSNE), Recurrent Autoencoder (RAE) and Variational-Recurrent Autoencoder (VRAE). We study two different classifiers for this task: neural networks and Support Vector Machines (SVMs). Furthermore, we investigate three query strategies, namely entropy, margin and random. 翻訳 |