LiteVLoc: Map-Lite Visual Localization for
Image Goal Navigation

Jianhao Jiao Jinhao He Changkun Liu Sebastian Aegidius Xiangcheng Hu
Tristan Braud Dimitrios Kanoulas

Abstract

This paper presents LiteVLoc, a hierarchical visual localization framework that uses a lightweight topo-metric map to represent the environment. The method consists of three sequential modules that estimate camera poses in a coarse-to-fine manner. Unlike mainstream approaches relying on detailed 3D representations, LiteVLoc reduces storage overhead by leveraging learning-based feature matching and geometric solvers for metric pose estimation. A novel dataset for the map-free relocalization task is also introduced. Extensive experiments including localization and navigation in both simulated and real-world scenarios have validate the system's performance and demonstrated its precision and efficiency for large-scale deployment.

Video

Evaluation on the Dataset for Map-Free Relocalization

We evaluate 13 state-of-the-art image matching methods under severe viewpoint changes, occlusions, and lighting variations. This will serve as an important experiment for our work. More results can be found in the paper and this page.

Visual Localization Evaluation

We evaluate the proposed LiteVLoc on 9 Simulated Sequences

We evaluate the proposed LiteVLoc on 3 Real-World Sequences

Cloud-Loop Visual Navigation on Simulated Environments

Left to Right: Goal Image - Feature Matching Result - Visual Observation (Color and Depth Image)

Red Trajectory: Estimated Trajectory. Green Trajectory: Planned Trajectory

Cloud-Loop Visual Navigation on Real-World Environments

Real-World Deployment

Environment 0 (Small Scale)

Goal Images are Given Sequentially

Environment 1 (Larger Scale)

Goal Images are Given Sequentially

Snippets

Manual Intervention

Obstacle Avoidance

Failure: Collision Because the Camera Cannot Recover Depth of Short-Distance Obstacles

BibTeX


      @article{jiao2024litevloc,
      title     = {LiteVLoc: Map-Lite Visual Localization for Image Goal Navigation},
      author    = {Jianhao Jiao, Jinhao He, Changkun Liu, Sebastian Aegidius, Xiangcheng Hu, Tristan Braud, Dimitrios Kanoulas},
      journal   = {arXiv preprint arXiv:xx},
      year      = {2024}
}