We launched this project to achieve UAV GeoLocalization in GPS-denied environment (GPS weak, unstable or unavailable). We used DJI Mavic air 2 for data collection and test flight. Other drone brands may lead to incorrect results for unknown reason.
- Google satellite map download, please refer to Deniz's contributed repo (2021-05)
- Figure out feature matching and apply SuperGlue as the matching algorithm among google satellite map and UAV taken image (2021-07)
- UAV flight rotation estimation (2022-03)
- UAV flight height estimation (2022-04)
- OSU campus flight test using Litchi (2022-04)
🚩 Updates
- Check out QuadTree, a spatial indexing algorithm that improves geo-queries in a 2D-space.
- Generate GIS building mask in correspondence with target area satellite image. GIS building mask comes from OSMnx could help improve geolocalization accuracy and UAV flight height estimation.
This is a vision-based project. We use images taken by UAV embedded camera as the only data source for geolocalization. Our completing method is to match features from UAV taken images with other data sources with similar contexture information such as GoogleSatelliteMap and OpenStreetMap. Therefore, we provide two sub-repos with respect to satellite map generation and corresponding GIS mask generation. Both repos require creating new conda environment due to specific libraries version dependencies.
See UAV Geolocalization for more details.
- Discussions or questions are welcomed. Please contact wei.909@osu.edu
- Our test flight is done around Ohio State University main campus. If you want test around other place, please recollect satellite image, GIS mask and rebuild featurebase.