VINet : Visual-inertial odometry as a sequence-to-sequence learning problem
- Submitting institution
-
The University of Warwick
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 6134
- Type
- E - Conference contribution
- DOI
-
-
- Title of conference / published proceedings
- Thirty-First AAAI Conference on Artificial Intelligence
- First page
- 3995
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- February
- Year of publication
- 2017
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
4
- Research group(s)
-
I - Artificial Intelligence and Human-Centred Computing
- Citation count
- 58
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Appearing at one of the two top conferences in artificial intelligence, this work introduces the first end-to-end trainable approach for visual-inertial ego-motion estimation, which is a fundamental building block in many robotics applications. It demonstrates that the expensive process of manual calibration between visual and inertial sensors can be dispensed by the deep learning framework. The new techniques in this paper have been adopted and extended e.g. for re-localization (Trigoni, Oxford), unsupervised estimation (Gu, Essex), and dense monocular SLAM (Davison, Imperial). The paper has also led to invited talks at major companies in this area including DJI, TuSimple and Samsung.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -