VidLoc: A Deep Spatio-Temporal Model for 6-DoF Videoclip Relocalization

Ronald Clark; Sen Wang; Andrew Markham; Niki Trigoni; Hongkai Wen

VidLoc: A Deep Spatio-Temporal Model for 6-DoF Videoclip Relocalization

Ronald Clark, Sen Wang, Andrew Markham, Niki Trigoni, Hongkai Wen

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

102 Downloads (Pure)

Abstract

Machine learning techniques, namely convolutional neural networks (CNN) and regression forests, have recently shown great promise in performing 6-DoF localization of monocular images. However, in most cases image-sequences, rather only single images, are readily available. To this extent, none of the proposed learning-based approaches exploit the valuable constraint of temporal smoothness, often leading to situations where the per-frame error is larger than the camera motion. In this paper we propose a recurrent model for performing 6-DoF localization of video-clips. We find that, even by considering only short sequences (20 frames), the pose estimates are smoothed and the localization error can be drastically reduced. Finally, we consider means of obtaining probabilistic pose estimates from our model. We evaluate our method on openly-available real-world autonomous driving and indoor localization datasets.

Original language	English
Title of host publication	2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
Publisher	IEEE
Pages	1-9
Number of pages	9
Publication status	Published - 21 Jul 2017

Access to Document

Download pdf
© 2017 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Accepted author manuscript, 4.47 MBLicence: All Rights Reserved

Cite this

@inproceedings{c73502f5393f46d8bb35672ef872970a,

title = "VidLoc: A Deep Spatio-Temporal Model for 6-DoF Videoclip Relocalization",

abstract = "Machine learning techniques, namely convolutional neural networks (CNN) and regression forests, have recently shown great promise in performing 6-DoF localization of monocular images. However, in most cases image-sequences, rather only single images, are readily available. To this extent, none of the proposed learning-based approaches exploit the valuable constraint of temporal smoothness, often leading to situations where the per-frame error is larger than the camera motion. In this paper we propose a recurrent model for performing 6-DoF localization of video-clips. We find that, even by considering only short sequences (20 frames), the pose estimates are smoothed and the localization error can be drastically reduced. Finally, we consider means of obtaining probabilistic pose estimates from our model. We evaluate our method on openly-available real-world autonomous driving and indoor localization datasets.",

author = "Ronald Clark and Sen Wang and Andrew Markham and Niki Trigoni and Hongkai Wen",

year = "2017",

month = jul,

day = "21",

language = "English",

pages = "1--9",

booktitle = "2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)",

publisher = "IEEE",

address = "United States",

}

TY - GEN

T1 - VidLoc: A Deep Spatio-Temporal Model for 6-DoF Videoclip Relocalization

AU - Clark, Ronald

AU - Wang, Sen

AU - Markham, Andrew

AU - Trigoni, Niki

AU - Wen, Hongkai

PY - 2017/7/21

Y1 - 2017/7/21

N2 - Machine learning techniques, namely convolutional neural networks (CNN) and regression forests, have recently shown great promise in performing 6-DoF localization of monocular images. However, in most cases image-sequences, rather only single images, are readily available. To this extent, none of the proposed learning-based approaches exploit the valuable constraint of temporal smoothness, often leading to situations where the per-frame error is larger than the camera motion. In this paper we propose a recurrent model for performing 6-DoF localization of video-clips. We find that, even by considering only short sequences (20 frames), the pose estimates are smoothed and the localization error can be drastically reduced. Finally, we consider means of obtaining probabilistic pose estimates from our model. We evaluate our method on openly-available real-world autonomous driving and indoor localization datasets.

AB - Machine learning techniques, namely convolutional neural networks (CNN) and regression forests, have recently shown great promise in performing 6-DoF localization of monocular images. However, in most cases image-sequences, rather only single images, are readily available. To this extent, none of the proposed learning-based approaches exploit the valuable constraint of temporal smoothness, often leading to situations where the per-frame error is larger than the camera motion. In this paper we propose a recurrent model for performing 6-DoF localization of video-clips. We find that, even by considering only short sequences (20 frames), the pose estimates are smoothed and the localization error can be drastically reduced. Finally, we consider means of obtaining probabilistic pose estimates from our model. We evaluate our method on openly-available real-world autonomous driving and indoor localization datasets.

M3 - Conference contribution

SP - 1

EP - 9

BT - 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

PB - IEEE

ER -

VidLoc: A Deep Spatio-Temporal Model for 6-DoF Videoclip Relocalization

Abstract

Access to Document

Fingerprint

Cite this