HANBAIKI.FYI is part documentary project, part educational exercise. The idea to record and map some of the thousands of vending machines (自動販売機・jidou hanbaiki) in Japan was there for a while. However, there is no way a single person could ever get around to do this on their own. So, I decided to crowdsource it. At the same time, I wanted to toy around with some new technology, including image recognition, that would help with sanitization of people's submissions.
This repository contains the source code for the Django project running HANBAIKI.FYI, managing uploading, processing, and display of submitted images. Uploaded photos are processed as follows:
- Analysis: We read EXIF data embedded into the image by the camera to extract the GPS coordinates of the subject. If those GPS coordinates could be read successfully, processing continues.
- Recognition: We use image recognition to check if uploaded files indeed contain vending machines. If the confidence score for an image lies above a certain threshold, processing continues.
- Tagging: We use text recognition to check if uploaded files contain any known Tags (e.g., beverage brands). If the confidence score for a given tag lies above a certain threshold, the tags are automatically associated to the uploaded vending machine.
- Publication: Processed images are published on our site alongside their GPS coordinates and basic location information, such as state (県・都), city (区・市), postcode (〒), and town (町・村), if available.