Is it possible to geographically offset imagery that is served via the TMS?

[spwoodcock]

Problem - satellite data alignment

• Often satellite imagery is slightly misaligned in the georeference
• This is quite common for Bing imagery used for OSM mapping, for example
• What should happen:
  • The imagery is offset to a known georeferenced image / map data
  • The OSM mapping commences with the correctly georeferenced imagery
• What happens in reality:
  • Either the user doesn't know the imagery is offset in the first place,
  • Or they don't know of a good dataset to georeference against to correct / offset
  • This means that (e.g. during a mapathon) a large area of data collected may actually have a large offset in OpenStreetMap
• This leaves us with a dilemma: once the offset is determined, do we:
  1. Correct the imagery offset, so all future data is correctly placed
  2. Continue to map data in the incorrect place, for consistency across the region
• In most cases, it is preferred to have data consistency then a mix of correctly and incorrectly aligned data, which is harder to pick apart later.
• Data sources such as drone imagery rarely have this problem, with very accurate horizontal georeferencing

Possible solution

• When we are mapping the same region again, perhaps with new data such as high-resolution recent drone imagery, we should adhere to the same offset as the original imagery (then perhaps in future the whole area can be corrected with a single offset shift of the data?)
• In order to do this, we need to serve the imagery to the user with a slight offset (so they can map aligned to the original offset and fill in gaps / correctly identify which features need to be modified on the map)
• What would really help in TiTiler is the ability to add a geographic offset at a specific bearing, perhaps via URL param on the TMS
• First off, any ideas how this could be done? Second, would a PR be welcomed to implement this, if we can settle on a potential solution?

Note

I work for HOTOSM, and we currently host OpenAerialMap. The latest version uses eoAPI, including TiTiler as a TMS server.

We load the TMS into tools like Tasking Manager, or uMap, so that people can map geometries remotely.

However, in order to map they need the offset set in the imagery.

[vincentsarago]

thanks for your question @spwoodcock

I have no real idea how this would be possible to be honest. I also believe that the offset would need to be adjusted by zoom level or to be set for the data raw resolution and then calculated for each served resolution.

The first easy think might be to create VRT files for your data, updating the GeoTransform, or maybe applying this offset directly on the client 🤷

[spwoodcock]

Thanks, this helps make it clearer for me! Its a tricky problem when we consider multiple data sources feeding into a TMS... perhaps it can only be achieved when a single COG dataset is selected

[SColchester]

Drone imagery might often be correctly aligned to ground control points or to WGS84, however from experience aligment is particularly an issue with drone imagery - because mappers don't look for alignment to ground control points or WGS84, they look for alignment first to existing mapped OSM data, and then existing Bing/Esri imagery - which although might be technically misaligned, make up the baseline for the data. So for drone imagery projets we tell mappers to manually alter the offset of the drone imagery to match existing data before they start mapping.

[vincentsarago]

Side note on GPCS + imagery:

in https://github.com/developmentseed/titiler-image we built a Reader (https://github.com/developmentseed/titiler-image/blob/main/titiler/image/reader.py) that accepted external GCPS to align any images

ref https://github.com/developmentseed/titiler-image/blob/main/docs/notebooks/titiler_image_GCPSReader.ipynb

[spwoodcock]

Sounds like a combination of solutions might be useful then:

1. Offset specified on OAM upload, if it is known so we can reproject / correct misaligned drone imagery (preferred)
2. Ability to offset via TMS (TiTiler) for imagery that already exists in OAM (if the offset wasn't known in advance).
3. Offset on the client side for misaligned satellite imagery / external sources that we don't control.

If its very difficult to offset via TiTiler, then perhaps client side offsetting would be the easiest - need to look into it further 👍

[SColchester]

But in short, if offset could be controlled on TMS side, particularly for custom imagery from OAM this would be a great step in the right direction for custom imagery - it would not fix everything but would at least give the mappers a much better starting point when using custom imagery.

[vincentsarago]

you could technically create a custom TMS which has vertical/horizontal offset e.g

import morecantile
from pyproj import CRS

custom_tms = morecantile.TileMatrixSet.custom(
    [-180, -85.0511287798066, 180, 85.0511287798066],
    corner_of_origin="topLeft",
    # Regular POO is `-20037508.342789244, 20037508.342789244`
    # Shifted POO by 1 Meter to the east and 1 Meter to South
    point_of_origin=[-20037507.342789244, 20037507.342789244],
    crs=CRS.from_epsg("3857"),
    extent_crs=CRS.from_epsg("4326"),
)

☝️ this would create a shifted WebMercatorQuad TMS

Another solution would be to have h/v_offset query parameter which could be forwarded to the Reader.tile() method to apply to the bbox for a given TMS

https://github.com/cogeotiff/rio-tiler/blob/ebacf0aeee47f44c2f321bd0a362212a374813ee/rio_tiler/io/rasterio.py#L285-L288

if you can share an image, I could make create an example notebook

[spwoodcock]

I believe @emi420 has this request too, so may have a good dataset for testing 🙏

Or perhaps @SColchester might too!