Contact: Lukas von Stumberg, Patrick Wenzel, Nan Yang, Prof. Daniel Cremers

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 in terms of robustness.

The paper can be downloaded at: https://arxiv.org/pdf/2010.06323
The relocalization tracking benchmark dataset first presented in our prior work GN-Net can be downloaded at: gnnet_benchmark_v1.4.zip
The supplementary can be downloaded at: lm-reloc-2020_supplementary.pdf
Code related to our benchmark can be found at: https://github.com/Artisense-ai/GN-Net-Benchmark

See also our previous work GN-Net, including the relocalization tracking benchmark used for evaluation in this work.

Export as PDF, XML, TEX or BIB

Conference and Workshop Papers
2020
[]		LM-Reloc: Levenberg-Marquardt Based Direct Visual Relocalization (L. von Stumberg, P. Wenzel, N. Yang and D. Cremers), In International Conference on 3D Vision (3DV), 2020. ([arXiv][project page][video][supplementary][poster]) [bibtex]