Skip to content

End-to-End Overview

This is an overview of the proposed end-to-end workflow that ties together all of HOT's tools.

Outcome: to enable communities to generate their own maps for their area of interest (neighbourhood, city, region), rich in data derived from local knowledge.

Requirements

  • A project owner / coordinator who can pull all of the pieces together.
  • Users in the area of interest with drones, and willing to produce imagery.
    • This has typically been a barrier, but with decent cheap consumer drones being available now, the opportunity to produce high-resolution base imagery has opened up.
  • A small team of digitisers, who use TM to generate a traning dataset for AI models.
    • This could feasibly be remote, but ideally local for better knowledge of what the reality is on the ground.
  • A group of local people who can verify features on the ground, adding tags to them with their local knowledge, via an easy to fill out mobile survey.

Methodology (Diagram)

Methodology (Description)

🚧 denotes 'blockers' or unknowns in the current workflow.

1. Collect Drone Imagery

  • Fly drones in grids to collect imagery for the entire area.
  • This will use Drone Aerial Tasking Manager (DATM), if it comes to fruition.

🚧 If DATM succeeds.

2. Process a Base Map & Upload to OAM

  • Get the imagery from the drones, and preprocess on a drive using EXIF info.
  • Use Node Open Drone Map (NodeODM) to merge into an mosaic.
  • Upload the mosaic to OAM.
  • A TMS URL is generated for you.

3. Digitise Features in Tasking Manager

  • Load the OAM imagery as a TMS into Tasking Manager.
  • Use a sandboxed version of TM for producing training data.
  • Only map a small sample area that is representative of the entire area.

🚧 We need an instance of TM isolated from OSM, so we don't mess with existing data. Alternatively we need a way to bulk delete geoms via iD Editor.

4. Generate a Model Using fAIr

  • Load the mapped features into fAIr and generate a model.
  • Use the model to predict the remaining features in the area.

5. Validate Predicted Features in TM

  • Load the predicted features back into a TM project.
    • fAIr allows for feedback to be given to the model, but this cannot be done collaboratively.
  • Collaboratively split the geoms as needed, and provide feedback on the generated geoms.

🚧 We should assess if we need this step, or the input from a single experienced validator directly in fAIr would be good enough.

🚧 Can iD Editor be used to split buildings nicely? Is this a good workflow?

6. Process Final Features in fAIr

  • Load the corrected features in fAIr.
  • The data could be used to re-train the model.
    • This process of fAIr-->TM-->fAIr could be done multiple times.
  • Output the final validated features as a GeoJSON.

7. Input Into FMTM

  • Now we have validated, AI-generated features for an area, they can be field validated and tagged.

    Note the features are not yet in OSM, as they will be field validated first.

  • As part of a field (mobile) survey, the features will be validated and tagged with useful information.
  • The final validated and tagged geometries will be bulk uploaded to OSM.

8. Extract from OpenSreetMap

  • Now final validation from FMTM is uploaded to OpenStreetMap , Now we need to download them in useful fileformats that we want
  • This will be achieved with HOT Export Tool which allows us to export the data we created in different file formats including ( Shpaefile , Geojson , KML , CSV , Flatgeobuff and manymore )

🚧 A way to feed validation failures back to TM for re-digitisation would be great.