Proof of concept, find GPS position based on one or more images

Given an input image, we find the closest image among a stored database of images. We then look up some meta data about it, and show it as well.

In the default example (see metadata.py, we show the GPS coordinates for each image.

This would enable you to for example detect where you are in the world by taking a picture ( assuming that you have a dataset of images with their corresponding GPS positions).

A slightly more complicated algorithm that might increase the accuracy in the GPS example would be something like:

• Take 5 different images of your surroundings.
• For each such image, find the top 10 nearest images.
• Find the GPS coordinates of all 50 images, then remove the 20 coordinates that are the furthest away from the rest.
• Take the average of the remaining 30 coordinates. This is where you are.

Usage:

Preparation

• Put image file you want to compare against in images/
• For each image file, set some metadata about them in metadata.py (update the metadata_lookup_map variable)
• Put the input image somewhere on disk.

Running

• Assuming you have poetry and Python 3 installed, install dependencies with poetry install
• Run the program: poetry run python imagesimilarity.py [path/to/your/input_image.jpg]

Example output

$ poetry run python imagesimilarity.py liberty2.jpg
Closest image is: images/liberty0.webp
Metadata: (40.6892, 74.0445) (gps coordinates)
similarity score: 0.9136244058609009