[2010.06323] LM-Reloc: Levenberg-Marquardt Based Direct Visual Relocalizationopen searchopen navigation menucontact arXivsubscribe to arXiv mailings

We present LM-Reloc -- a novel approach for visual relocalization based on direct image alignment. In contrast to prior works that tackle the problem with a feature-based formulation, the proposed method does not rely on feature matching and RANSAC. Hence, the method can utilize not only corners but any region of the image with gradients. In particular, we propose a loss formulation inspired by the classical Levenberg-Marquardt algorithm to train LM-Net. The learned features significantly improve the robustness of direct image alignment, especially for relocalization across different conditions. To further improve the robustness of LM-Net against large image baselines, we propose a pose estimation network, CorrPoseNet, which regresses the relative pose to bootstrap the direct image alignment. Evaluations on the CARLA and Oxford RobotCar relocalization tracking benchmark show that our approach delivers more accurate results than previous state-of-the-art methods while being comparable i

Date: 2020/10/18 03:51

Related Entries

Read more [1904.11932] GN-Net: The Gauss-Newton Loss for Multi-Weather Relocalizationcontact arXivarXiv Twitte...
0 users, 1 mentions 2020/02/10 15:01
Read more [2002.03983] StickyPillars: Robust feature matching on point clouds using Graph Neural Networksconta...
0 users, 1 mentions 2020/02/11 06:51
Read more Michael Bernstein · Stanford HCI
0 users, 1 mentions 2020/04/25 08:21
Read more SuperGlue CVPR 2020
0 users, 1 mentions 2020/06/22 14:21
Read more [2010.01086] Semi-Supervised Learning for Multi-Task Scene Understanding by Neural Graph Consensusop...
0 users, 1 mentions 2020/10/06 00:52