ViPR: Visual-Odometry-aided Pose Regression for 6DoF Camera Localization

  • Post published:October 17, 2020

Ott F., Feigl T., Löffler C., Mutschler C.:

In: The IEEE Computer Vision Foundation (CVF), Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, Joint Workshop on Long-Term Visual Localization, Visual Odometry and Geometric and Learning-based SLAM, Seattle, Washington, 2020.

Visual Odometry (VO) accumulates a positional drift in long-term robot navigation tasks. Although Convolutional Neural Networks (CNNs) improve VO in various aspects, VO still suffers from moving obstacles, discontinuous ob- servation of features, and poor textures or visual informa- tion. While recent approaches estimate a 6DoF pose ei- ther directly from (a series of) images or by merging depth maps with optical flow (OF), research that combines abso- lute pose regression with OF is limited. We propose ViPR, a novel modular architecture for long- term 6DoF VO that leverages temporal information and synergies between absolute pose estimates (from PoseNet- like modules) and relative pose estimates (from FlowNet- based modules) by combining both through recurrent lay- ers. Experiments on known datasets and on our own Indus- try dataset show that our modular design outperforms state of the art in long-term navigation tasks.

Open Access: http://openaccess.thecvf.com/content_CVPRW_2020/papers/w3/Ott_ViPR_Visual-Odometry-Aided_Pose_Regression_for_6DoF_Camera_Localization_CVPRW_2020_paper.pdf
URL: http://openaccess.thecvf.com/content_CVPRW_2020/html/w3/Ott_ViPR_Visual-Odometry-Aided_Pose_Regression_for_6DoF_Camera_Localization_CVPRW_2020_paper.html
BibTeX: Download