.png){kind=icon}
.png){kind=icon}
Return
A Novel UAV Visual Navigation Method Using Online Customizable Image Reference
.png){kind=icon}
PRE
AI
DOI：10.1109/LGRS.2024.3465464.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Scene matching-based navigation is a common method for unmanned aerial vehicle (UAV) visual navigation. However, static maps and limited memory onboard pose challenges in balancing localization precision and the effective area of localization for existing methods. To address these issues, this work generates customizable image references onboard and proposes a hierarchical method for scene matching-based navigation. The proposed localization method trains the instant neural radiance field (NeRF) to learn 3-D implicit representations offline and generate customizable image references online. Moreover, a coarse-to-fine strategy is employed to enhance the accuracy of localization. Retrieval maps and localization maps are established for the coarse and fine stages, respectively. The experiments demonstrate that our localization achieves a comparable RMSE of less than 1.5 m in virtual datasets and 3 m in real datasets, indicating its competitive performance.
Keywords:
Location awareness
Three-dimensional displays
Neural radiance field
Accuracy
Navigation
Rendering (computer graphics)
Visualization
3-D reconstruction
instant neural radiance field (NeRF)
online customizable
remote sensing image
similarity measurement
unmanned aerial vehicle (UAV) visual localization
Journal
IF:
16.4
Papers:
1.0W
Citations:
5.1K