Geo-location Prediction via Landmarks

This tool derives the GPS coordinates at which a video (in Delft) was taken.

It does this by extracting ORB features from the input video, and matching these to a labelled database of landmarks in Delft.

Example

1. new-church-delft.m4v
   

2. Running the script
   

3. Pasting 52°00'45.0612"N 4°21'39.2724"E in Google Maps
   

Using the tool

1. Install dependencies: $ pip install -r requirements.txt
2. Run the script: $ python src/main.py

Creating your own dataset

Requisites: a folder with images from landmarks.

NOTE: If an image in your dataset doesn't contain GPS metadata, the script will prompt you for manual provisioning. Set SKIP_UNKNOWN_GEO in src/img_index.py to True to instead skip said images.

Steps

1. Delete the index folder, which contains the default Delft dataset.
2. Rerun the script: $ python src/main.py, it will guide you through the creation of your own dataset.

License

This repository is released under the MIT license. See LICENSE for additional details.