Visual odometry based on convolutional neural networks for large-scale scenes

Xuyang Meng; Chunxiao Fan; Yue Ming

doi:10.1117/12.2524278

6 May 2019 Visual odometry based on convolutional neural networks for large-scale scenes

Xuyang Meng, Chunxiao Fan, Yue Ming

Proceedings Volume 11069, Tenth International Conference on Graphics and Image Processing (ICGIP 2018); 110690A (2019) https://doi.org/10.1117/12.2524278
Event: Tenth International Conference on Graphic and Image Processing (ICGIP 2018), 2018, Chengdu, China

Abstract

The task of visual odometry (VO) is to estimation camera motion and image depth, which is the main part of 3D reconstruction and the front-end of simultaneous localization and mapping (SLAM). However, the accuracy of most of the existing methods is low or some advanced sensors are required. In order to predict camera pose and image depth at the same time with high accuracy from image sequences captured by monocular camera, we train a novel framework, named MD-Net, and it is based on convolutional neural networks (CNNs). There are two main modules: one is camera motion estimator which is able to estimate the 6-DoF camera pose, the other is depth estimator computing the depth of its view. The keys of our proposed framework are that we can not only train our two independent estimators, but also predict depth and camera motion simultaneously. What’s more, our motion estimator includes some shared convolutional layers and is divided into two branches to estimate camera orientation and translation, respectively. Experiments on KITTI dataset and TUM dataset show that our proposed method can extract meaningful depth estimation and successfully estimate frame-to-frame camera rotations and translations in large scenes even texture-less. It outperforms previous methods in terms of accuracy and robustness.

Citation Download Citation

Xuyang Meng, Chunxiao Fan, and Yue Ming "Visual odometry based on convolutional neural networks for large-scale scenes", Proc. SPIE 11069, Tenth International Conference on Graphics and Image Processing (ICGIP 2018), 110690A (6 May 2019); https://doi.org/10.1117/12.2524278

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available